1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * The User Datagram Protocol (UDP).
8 *
9 * Authors: Ross Biro
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
12 * Alan Cox, <alan@lxorguk.ukuu.org.uk>
13 * Hirokazu Takahashi, <taka@valinux.co.jp>
14 *
15 * Fixes:
16 * Alan Cox : verify_area() calls
17 * Alan Cox : stopped close while in use off icmp
18 * messages. Not a fix but a botch that
19 * for udp at least is 'valid'.
20 * Alan Cox : Fixed icmp handling properly
21 * Alan Cox : Correct error for oversized datagrams
22 * Alan Cox : Tidied select() semantics.
23 * Alan Cox : udp_err() fixed properly, also now
24 * select and read wake correctly on errors
25 * Alan Cox : udp_send verify_area moved to avoid mem leak
26 * Alan Cox : UDP can count its memory
27 * Alan Cox : send to an unknown connection causes
28 * an ECONNREFUSED off the icmp, but
29 * does NOT close.
30 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
31 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
32 * bug no longer crashes it.
33 * Fred Van Kempen : Net2e support for sk->broadcast.
34 * Alan Cox : Uses skb_free_datagram
35 * Alan Cox : Added get/set sockopt support.
36 * Alan Cox : Broadcasting without option set returns EACCES.
37 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
38 * Alan Cox : Use ip_tos and ip_ttl
39 * Alan Cox : SNMP Mibs
40 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
41 * Matt Dillon : UDP length checks.
42 * Alan Cox : Smarter af_inet used properly.
43 * Alan Cox : Use new kernel side addressing.
44 * Alan Cox : Incorrect return on truncated datagram receive.
45 * Arnt Gulbrandsen : New udp_send and stuff
46 * Alan Cox : Cache last socket
47 * Alan Cox : Route cache
48 * Jon Peatfield : Minor efficiency fix to sendto().
49 * Mike Shaver : RFC1122 checks.
50 * Alan Cox : Nonblocking error fix.
51 * Willy Konynenberg : Transparent proxying support.
52 * Mike McLagan : Routing by source
53 * David S. Miller : New socket lookup architecture.
54 * Last socket cache retained as it
55 * does have a high hit rate.
56 * Olaf Kirch : Don't linearise iovec on sendmsg.
57 * Andi Kleen : Some cleanups, cache destination entry
58 * for connect.
59 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
60 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
61 * return ENOTCONN for unconnected sockets (POSIX)
62 * Janos Farkas : don't deliver multi/broadcasts to a different
63 * bound-to-device socket
64 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * datagrams.
66 * Hirokazu Takahashi : sendfile() on UDP works now.
67 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
68 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
69 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
70 * a single port at the same time.
71 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
72 * James Chapman : Add L2TP encapsulation type.
73 */
74
75 #define pr_fmt(fmt) "UDP: " fmt
76
77 #include <linux/bpf-cgroup.h>
78 #include <linux/uaccess.h>
79 #include <asm/ioctls.h>
80 #include <linux/memblock.h>
81 #include <linux/highmem.h>
82 #include <linux/types.h>
83 #include <linux/fcntl.h>
84 #include <linux/module.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/igmp.h>
88 #include <linux/inetdevice.h>
89 #include <linux/in.h>
90 #include <linux/errno.h>
91 #include <linux/timer.h>
92 #include <linux/mm.h>
93 #include <linux/inet.h>
94 #include <linux/netdevice.h>
95 #include <linux/slab.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/inet_hashtables.h>
103 #include <net/ip_tunnels.h>
104 #include <net/route.h>
105 #include <net/checksum.h>
106 #include <net/gso.h>
107 #include <net/xfrm.h>
108 #include <trace/events/udp.h>
109 #include <linux/static_key.h>
110 #include <linux/btf_ids.h>
111 #include <trace/events/skb.h>
112 #include <net/busy_poll.h>
113 #include "udp_impl.h"
114 #include <net/sock_reuseport.h>
115 #include <net/addrconf.h>
116 #include <net/udp_tunnel.h>
117 #include <net/gro.h>
118 #include <net/inet_dscp.h>
119 #if IS_ENABLED(CONFIG_IPV6)
120 #include <net/ipv6_stubs.h>
121 #endif
122
123 struct udp_table udp_table __read_mostly;
124 EXPORT_SYMBOL(udp_table);
125
126 long sysctl_udp_mem[3] __read_mostly;
127 EXPORT_SYMBOL(sysctl_udp_mem);
128
129 atomic_long_t udp_memory_allocated ____cacheline_aligned_in_smp;
130 EXPORT_SYMBOL(udp_memory_allocated);
131 DEFINE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
132 EXPORT_PER_CPU_SYMBOL_GPL(udp_memory_per_cpu_fw_alloc);
133
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN_PERNET)
136
udp_get_table_prot(struct sock * sk)137 static struct udp_table *udp_get_table_prot(struct sock *sk)
138 {
139 return sk->sk_prot->h.udp_table ? : sock_net(sk)->ipv4.udp_table;
140 }
141
udp_lib_lport_inuse(struct net * net,__u16 num,const struct udp_hslot * hslot,unsigned long * bitmap,struct sock * sk,unsigned int log)142 static int udp_lib_lport_inuse(struct net *net, __u16 num,
143 const struct udp_hslot *hslot,
144 unsigned long *bitmap,
145 struct sock *sk, unsigned int log)
146 {
147 struct sock *sk2;
148 kuid_t uid = sock_i_uid(sk);
149
150 sk_for_each(sk2, &hslot->head) {
151 if (net_eq(sock_net(sk2), net) &&
152 sk2 != sk &&
153 (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154 (!sk2->sk_reuse || !sk->sk_reuse) &&
155 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157 inet_rcv_saddr_equal(sk, sk2, true)) {
158 if (sk2->sk_reuseport && sk->sk_reuseport &&
159 !rcu_access_pointer(sk->sk_reuseport_cb) &&
160 uid_eq(uid, sock_i_uid(sk2))) {
161 if (!bitmap)
162 return 0;
163 } else {
164 if (!bitmap)
165 return 1;
166 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167 bitmap);
168 }
169 }
170 }
171 return 0;
172 }
173
174 /*
175 * Note: we still hold spinlock of primary hash chain, so no other writer
176 * can insert/delete a socket with local_port == num
177 */
udp_lib_lport_inuse2(struct net * net,__u16 num,struct udp_hslot * hslot2,struct sock * sk)178 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179 struct udp_hslot *hslot2,
180 struct sock *sk)
181 {
182 struct sock *sk2;
183 kuid_t uid = sock_i_uid(sk);
184 int res = 0;
185
186 spin_lock(&hslot2->lock);
187 udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188 if (net_eq(sock_net(sk2), net) &&
189 sk2 != sk &&
190 (udp_sk(sk2)->udp_port_hash == num) &&
191 (!sk2->sk_reuse || !sk->sk_reuse) &&
192 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194 inet_rcv_saddr_equal(sk, sk2, true)) {
195 if (sk2->sk_reuseport && sk->sk_reuseport &&
196 !rcu_access_pointer(sk->sk_reuseport_cb) &&
197 uid_eq(uid, sock_i_uid(sk2))) {
198 res = 0;
199 } else {
200 res = 1;
201 }
202 break;
203 }
204 }
205 spin_unlock(&hslot2->lock);
206 return res;
207 }
208
udp_reuseport_add_sock(struct sock * sk,struct udp_hslot * hslot)209 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 {
211 struct net *net = sock_net(sk);
212 kuid_t uid = sock_i_uid(sk);
213 struct sock *sk2;
214
215 sk_for_each(sk2, &hslot->head) {
216 if (net_eq(sock_net(sk2), net) &&
217 sk2 != sk &&
218 sk2->sk_family == sk->sk_family &&
219 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220 (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221 (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223 inet_rcv_saddr_equal(sk, sk2, false)) {
224 return reuseport_add_sock(sk, sk2,
225 inet_rcv_saddr_any(sk));
226 }
227 }
228
229 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
230 }
231
232 /**
233 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
234 *
235 * @sk: socket struct in question
236 * @snum: port number to look up
237 * @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238 * with NULL address
239 */
udp_lib_get_port(struct sock * sk,unsigned short snum,unsigned int hash2_nulladdr)240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241 unsigned int hash2_nulladdr)
242 {
243 struct udp_table *udptable = udp_get_table_prot(sk);
244 struct udp_hslot *hslot, *hslot2;
245 struct net *net = sock_net(sk);
246 int error = -EADDRINUSE;
247
248 if (!snum) {
249 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
250 unsigned short first, last;
251 int low, high, remaining;
252 unsigned int rand;
253
254 inet_sk_get_local_port_range(sk, &low, &high);
255 remaining = (high - low) + 1;
256
257 rand = get_random_u32();
258 first = reciprocal_scale(rand, remaining) + low;
259 /*
260 * force rand to be an odd multiple of UDP_HTABLE_SIZE
261 */
262 rand = (rand | 1) * (udptable->mask + 1);
263 last = first + udptable->mask + 1;
264 do {
265 hslot = udp_hashslot(udptable, net, first);
266 bitmap_zero(bitmap, PORTS_PER_CHAIN);
267 spin_lock_bh(&hslot->lock);
268 udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
269 udptable->log);
270
271 snum = first;
272 /*
273 * Iterate on all possible values of snum for this hash.
274 * Using steps of an odd multiple of UDP_HTABLE_SIZE
275 * give us randomization and full range coverage.
276 */
277 do {
278 if (low <= snum && snum <= high &&
279 !test_bit(snum >> udptable->log, bitmap) &&
280 !inet_is_local_reserved_port(net, snum))
281 goto found;
282 snum += rand;
283 } while (snum != first);
284 spin_unlock_bh(&hslot->lock);
285 cond_resched();
286 } while (++first != last);
287 goto fail;
288 } else {
289 hslot = udp_hashslot(udptable, net, snum);
290 spin_lock_bh(&hslot->lock);
291 if (hslot->count > 10) {
292 int exist;
293 unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
294
295 slot2 &= udptable->mask;
296 hash2_nulladdr &= udptable->mask;
297
298 hslot2 = udp_hashslot2(udptable, slot2);
299 if (hslot->count < hslot2->count)
300 goto scan_primary_hash;
301
302 exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
303 if (!exist && (hash2_nulladdr != slot2)) {
304 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
305 exist = udp_lib_lport_inuse2(net, snum, hslot2,
306 sk);
307 }
308 if (exist)
309 goto fail_unlock;
310 else
311 goto found;
312 }
313 scan_primary_hash:
314 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
315 goto fail_unlock;
316 }
317 found:
318 inet_sk(sk)->inet_num = snum;
319 udp_sk(sk)->udp_port_hash = snum;
320 udp_sk(sk)->udp_portaddr_hash ^= snum;
321 if (sk_unhashed(sk)) {
322 if (sk->sk_reuseport &&
323 udp_reuseport_add_sock(sk, hslot)) {
324 inet_sk(sk)->inet_num = 0;
325 udp_sk(sk)->udp_port_hash = 0;
326 udp_sk(sk)->udp_portaddr_hash ^= snum;
327 goto fail_unlock;
328 }
329
330 sock_set_flag(sk, SOCK_RCU_FREE);
331
332 sk_add_node_rcu(sk, &hslot->head);
333 hslot->count++;
334 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
335
336 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
337 spin_lock(&hslot2->lock);
338 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
339 sk->sk_family == AF_INET6)
340 hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
341 &hslot2->head);
342 else
343 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
344 &hslot2->head);
345 hslot2->count++;
346 spin_unlock(&hslot2->lock);
347 }
348
349 error = 0;
350 fail_unlock:
351 spin_unlock_bh(&hslot->lock);
352 fail:
353 return error;
354 }
355 EXPORT_SYMBOL(udp_lib_get_port);
356
udp_v4_get_port(struct sock * sk,unsigned short snum)357 int udp_v4_get_port(struct sock *sk, unsigned short snum)
358 {
359 unsigned int hash2_nulladdr =
360 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
361 unsigned int hash2_partial =
362 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
363
364 /* precompute partial secondary hash */
365 udp_sk(sk)->udp_portaddr_hash = hash2_partial;
366 return udp_lib_get_port(sk, snum, hash2_nulladdr);
367 }
368
compute_score(struct sock * sk,const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned short hnum,int dif,int sdif)369 static int compute_score(struct sock *sk, const struct net *net,
370 __be32 saddr, __be16 sport,
371 __be32 daddr, unsigned short hnum,
372 int dif, int sdif)
373 {
374 int score;
375 struct inet_sock *inet;
376 bool dev_match;
377
378 if (!net_eq(sock_net(sk), net) ||
379 udp_sk(sk)->udp_port_hash != hnum ||
380 ipv6_only_sock(sk))
381 return -1;
382
383 if (sk->sk_rcv_saddr != daddr)
384 return -1;
385
386 score = (sk->sk_family == PF_INET) ? 2 : 1;
387
388 inet = inet_sk(sk);
389 if (inet->inet_daddr) {
390 if (inet->inet_daddr != saddr)
391 return -1;
392 score += 4;
393 }
394
395 if (inet->inet_dport) {
396 if (inet->inet_dport != sport)
397 return -1;
398 score += 4;
399 }
400
401 dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
402 dif, sdif);
403 if (!dev_match)
404 return -1;
405 if (sk->sk_bound_dev_if)
406 score += 4;
407
408 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
409 score++;
410 return score;
411 }
412
413 INDIRECT_CALLABLE_SCOPE
udp_ehashfn(const struct net * net,const __be32 laddr,const __u16 lport,const __be32 faddr,const __be16 fport)414 u32 udp_ehashfn(const struct net *net, const __be32 laddr, const __u16 lport,
415 const __be32 faddr, const __be16 fport)
416 {
417 net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
418
419 return __inet_ehashfn(laddr, lport, faddr, fport,
420 udp_ehash_secret + net_hash_mix(net));
421 }
422
423 /* called with rcu_read_lock() */
udp4_lib_lookup2(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,unsigned int hnum,int dif,int sdif,struct udp_hslot * hslot2,struct sk_buff * skb)424 static struct sock *udp4_lib_lookup2(const struct net *net,
425 __be32 saddr, __be16 sport,
426 __be32 daddr, unsigned int hnum,
427 int dif, int sdif,
428 struct udp_hslot *hslot2,
429 struct sk_buff *skb)
430 {
431 struct sock *sk, *result;
432 int score, badness;
433 bool need_rescore;
434
435 result = NULL;
436 badness = 0;
437 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
438 need_rescore = false;
439 rescore:
440 score = compute_score(need_rescore ? result : sk, net, saddr,
441 sport, daddr, hnum, dif, sdif);
442 if (score > badness) {
443 badness = score;
444
445 if (need_rescore)
446 continue;
447
448 if (sk->sk_state == TCP_ESTABLISHED) {
449 result = sk;
450 continue;
451 }
452
453 result = inet_lookup_reuseport(net, sk, skb, sizeof(struct udphdr),
454 saddr, sport, daddr, hnum, udp_ehashfn);
455 if (!result) {
456 result = sk;
457 continue;
458 }
459
460 /* Fall back to scoring if group has connections */
461 if (!reuseport_has_conns(sk))
462 return result;
463
464 /* Reuseport logic returned an error, keep original score. */
465 if (IS_ERR(result))
466 continue;
467
468 /* compute_score is too long of a function to be
469 * inlined, and calling it again here yields
470 * measureable overhead for some
471 * workloads. Work around it by jumping
472 * backwards to rescore 'result'.
473 */
474 need_rescore = true;
475 goto rescore;
476 }
477 }
478 return result;
479 }
480
481 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
482 * harder than this. -DaveM
483 */
__udp4_lib_lookup(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif,int sdif,struct udp_table * udptable,struct sk_buff * skb)484 struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
485 __be16 sport, __be32 daddr, __be16 dport, int dif,
486 int sdif, struct udp_table *udptable, struct sk_buff *skb)
487 {
488 unsigned short hnum = ntohs(dport);
489 unsigned int hash2, slot2;
490 struct udp_hslot *hslot2;
491 struct sock *result, *sk;
492
493 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
494 slot2 = hash2 & udptable->mask;
495 hslot2 = &udptable->hash2[slot2];
496
497 /* Lookup connected or non-wildcard socket */
498 result = udp4_lib_lookup2(net, saddr, sport,
499 daddr, hnum, dif, sdif,
500 hslot2, skb);
501 if (!IS_ERR_OR_NULL(result) && result->sk_state == TCP_ESTABLISHED)
502 goto done;
503
504 /* Lookup redirect from BPF */
505 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
506 udptable == net->ipv4.udp_table) {
507 sk = inet_lookup_run_sk_lookup(net, IPPROTO_UDP, skb, sizeof(struct udphdr),
508 saddr, sport, daddr, hnum, dif,
509 udp_ehashfn);
510 if (sk) {
511 result = sk;
512 goto done;
513 }
514 }
515
516 /* Got non-wildcard socket or error on first lookup */
517 if (result)
518 goto done;
519
520 /* Lookup wildcard sockets */
521 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
522 slot2 = hash2 & udptable->mask;
523 hslot2 = &udptable->hash2[slot2];
524
525 result = udp4_lib_lookup2(net, saddr, sport,
526 htonl(INADDR_ANY), hnum, dif, sdif,
527 hslot2, skb);
528 done:
529 if (IS_ERR(result))
530 return NULL;
531 return result;
532 }
533 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
534
__udp4_lib_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport,struct udp_table * udptable)535 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
536 __be16 sport, __be16 dport,
537 struct udp_table *udptable)
538 {
539 const struct iphdr *iph = ip_hdr(skb);
540
541 return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
542 iph->daddr, dport, inet_iif(skb),
543 inet_sdif(skb), udptable, skb);
544 }
545
udp4_lib_lookup_skb(const struct sk_buff * skb,__be16 sport,__be16 dport)546 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
547 __be16 sport, __be16 dport)
548 {
549 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
550 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
551 struct net *net = dev_net(skb->dev);
552 int iif, sdif;
553
554 inet_get_iif_sdif(skb, &iif, &sdif);
555
556 return __udp4_lib_lookup(net, iph->saddr, sport,
557 iph->daddr, dport, iif,
558 sdif, net->ipv4.udp_table, NULL);
559 }
560
561 /* Must be called under rcu_read_lock().
562 * Does increment socket refcount.
563 */
564 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
udp4_lib_lookup(const struct net * net,__be32 saddr,__be16 sport,__be32 daddr,__be16 dport,int dif)565 struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
566 __be32 daddr, __be16 dport, int dif)
567 {
568 struct sock *sk;
569
570 sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
571 dif, 0, net->ipv4.udp_table, NULL);
572 if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
573 sk = NULL;
574 return sk;
575 }
576 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
577 #endif
578
__udp_is_mcast_sock(struct net * net,const struct sock * sk,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif,unsigned short hnum)579 static inline bool __udp_is_mcast_sock(struct net *net, const struct sock *sk,
580 __be16 loc_port, __be32 loc_addr,
581 __be16 rmt_port, __be32 rmt_addr,
582 int dif, int sdif, unsigned short hnum)
583 {
584 const struct inet_sock *inet = inet_sk(sk);
585
586 if (!net_eq(sock_net(sk), net) ||
587 udp_sk(sk)->udp_port_hash != hnum ||
588 (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
589 (inet->inet_dport != rmt_port && inet->inet_dport) ||
590 (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
591 ipv6_only_sock(sk) ||
592 !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
593 return false;
594 if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
595 return false;
596 return true;
597 }
598
599 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
600 EXPORT_SYMBOL(udp_encap_needed_key);
601
602 #if IS_ENABLED(CONFIG_IPV6)
603 DEFINE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
604 EXPORT_SYMBOL(udpv6_encap_needed_key);
605 #endif
606
udp_encap_enable(void)607 void udp_encap_enable(void)
608 {
609 static_branch_inc(&udp_encap_needed_key);
610 }
611 EXPORT_SYMBOL(udp_encap_enable);
612
udp_encap_disable(void)613 void udp_encap_disable(void)
614 {
615 static_branch_dec(&udp_encap_needed_key);
616 }
617 EXPORT_SYMBOL(udp_encap_disable);
618
619 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
620 * through error handlers in encapsulations looking for a match.
621 */
__udp4_lib_err_encap_no_sk(struct sk_buff * skb,u32 info)622 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
623 {
624 int i;
625
626 for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
627 int (*handler)(struct sk_buff *skb, u32 info);
628 const struct ip_tunnel_encap_ops *encap;
629
630 encap = rcu_dereference(iptun_encaps[i]);
631 if (!encap)
632 continue;
633 handler = encap->err_handler;
634 if (handler && !handler(skb, info))
635 return 0;
636 }
637
638 return -ENOENT;
639 }
640
641 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
642 * reversing source and destination port: this will match tunnels that force the
643 * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
644 * lwtunnels might actually break this assumption by being configured with
645 * different destination ports on endpoints, in this case we won't be able to
646 * trace ICMP messages back to them.
647 *
648 * If this doesn't match any socket, probe tunnels with arbitrary destination
649 * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
650 * we've sent packets to won't necessarily match the local destination port.
651 *
652 * Then ask the tunnel implementation to match the error against a valid
653 * association.
654 *
655 * Return an error if we can't find a match, the socket if we need further
656 * processing, zero otherwise.
657 */
__udp4_lib_err_encap(struct net * net,const struct iphdr * iph,struct udphdr * uh,struct udp_table * udptable,struct sock * sk,struct sk_buff * skb,u32 info)658 static struct sock *__udp4_lib_err_encap(struct net *net,
659 const struct iphdr *iph,
660 struct udphdr *uh,
661 struct udp_table *udptable,
662 struct sock *sk,
663 struct sk_buff *skb, u32 info)
664 {
665 int (*lookup)(struct sock *sk, struct sk_buff *skb);
666 int network_offset, transport_offset;
667 struct udp_sock *up;
668
669 network_offset = skb_network_offset(skb);
670 transport_offset = skb_transport_offset(skb);
671
672 /* Network header needs to point to the outer IPv4 header inside ICMP */
673 skb_reset_network_header(skb);
674
675 /* Transport header needs to point to the UDP header */
676 skb_set_transport_header(skb, iph->ihl << 2);
677
678 if (sk) {
679 up = udp_sk(sk);
680
681 lookup = READ_ONCE(up->encap_err_lookup);
682 if (lookup && lookup(sk, skb))
683 sk = NULL;
684
685 goto out;
686 }
687
688 sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
689 iph->saddr, uh->dest, skb->dev->ifindex, 0,
690 udptable, NULL);
691 if (sk) {
692 up = udp_sk(sk);
693
694 lookup = READ_ONCE(up->encap_err_lookup);
695 if (!lookup || lookup(sk, skb))
696 sk = NULL;
697 }
698
699 out:
700 if (!sk)
701 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
702
703 skb_set_transport_header(skb, transport_offset);
704 skb_set_network_header(skb, network_offset);
705
706 return sk;
707 }
708
709 /*
710 * This routine is called by the ICMP module when it gets some
711 * sort of error condition. If err < 0 then the socket should
712 * be closed and the error returned to the user. If err > 0
713 * it's just the icmp type << 8 | icmp code.
714 * Header points to the ip header of the error packet. We move
715 * on past this. Then (as it used to claim before adjustment)
716 * header points to the first 8 bytes of the udp header. We need
717 * to find the appropriate port.
718 */
719
__udp4_lib_err(struct sk_buff * skb,u32 info,struct udp_table * udptable)720 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
721 {
722 struct inet_sock *inet;
723 const struct iphdr *iph = (const struct iphdr *)skb->data;
724 struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
725 const int type = icmp_hdr(skb)->type;
726 const int code = icmp_hdr(skb)->code;
727 bool tunnel = false;
728 struct sock *sk;
729 int harderr;
730 int err;
731 struct net *net = dev_net(skb->dev);
732
733 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
734 iph->saddr, uh->source, skb->dev->ifindex,
735 inet_sdif(skb), udptable, NULL);
736
737 if (!sk || READ_ONCE(udp_sk(sk)->encap_type)) {
738 /* No socket for error: try tunnels before discarding */
739 if (static_branch_unlikely(&udp_encap_needed_key)) {
740 sk = __udp4_lib_err_encap(net, iph, uh, udptable, sk, skb,
741 info);
742 if (!sk)
743 return 0;
744 } else
745 sk = ERR_PTR(-ENOENT);
746
747 if (IS_ERR(sk)) {
748 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
749 return PTR_ERR(sk);
750 }
751
752 tunnel = true;
753 }
754
755 err = 0;
756 harderr = 0;
757 inet = inet_sk(sk);
758
759 switch (type) {
760 default:
761 case ICMP_TIME_EXCEEDED:
762 err = EHOSTUNREACH;
763 break;
764 case ICMP_SOURCE_QUENCH:
765 goto out;
766 case ICMP_PARAMETERPROB:
767 err = EPROTO;
768 harderr = 1;
769 break;
770 case ICMP_DEST_UNREACH:
771 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
772 ipv4_sk_update_pmtu(skb, sk, info);
773 if (READ_ONCE(inet->pmtudisc) != IP_PMTUDISC_DONT) {
774 err = EMSGSIZE;
775 harderr = 1;
776 break;
777 }
778 goto out;
779 }
780 err = EHOSTUNREACH;
781 if (code <= NR_ICMP_UNREACH) {
782 harderr = icmp_err_convert[code].fatal;
783 err = icmp_err_convert[code].errno;
784 }
785 break;
786 case ICMP_REDIRECT:
787 ipv4_sk_redirect(skb, sk);
788 goto out;
789 }
790
791 /*
792 * RFC1122: OK. Passes ICMP errors back to application, as per
793 * 4.1.3.3.
794 */
795 if (tunnel) {
796 /* ...not for tunnels though: we don't have a sending socket */
797 if (udp_sk(sk)->encap_err_rcv)
798 udp_sk(sk)->encap_err_rcv(sk, skb, err, uh->dest, info,
799 (u8 *)(uh+1));
800 goto out;
801 }
802 if (!inet_test_bit(RECVERR, sk)) {
803 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
804 goto out;
805 } else
806 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
807
808 sk->sk_err = err;
809 sk_error_report(sk);
810 out:
811 return 0;
812 }
813
udp_err(struct sk_buff * skb,u32 info)814 int udp_err(struct sk_buff *skb, u32 info)
815 {
816 return __udp4_lib_err(skb, info, dev_net(skb->dev)->ipv4.udp_table);
817 }
818
819 /*
820 * Throw away all pending data and cancel the corking. Socket is locked.
821 */
udp_flush_pending_frames(struct sock * sk)822 void udp_flush_pending_frames(struct sock *sk)
823 {
824 struct udp_sock *up = udp_sk(sk);
825
826 if (up->pending) {
827 up->len = 0;
828 WRITE_ONCE(up->pending, 0);
829 ip_flush_pending_frames(sk);
830 }
831 }
832 EXPORT_SYMBOL(udp_flush_pending_frames);
833
834 /**
835 * udp4_hwcsum - handle outgoing HW checksumming
836 * @skb: sk_buff containing the filled-in UDP header
837 * (checksum field must be zeroed out)
838 * @src: source IP address
839 * @dst: destination IP address
840 */
udp4_hwcsum(struct sk_buff * skb,__be32 src,__be32 dst)841 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
842 {
843 struct udphdr *uh = udp_hdr(skb);
844 int offset = skb_transport_offset(skb);
845 int len = skb->len - offset;
846 int hlen = len;
847 __wsum csum = 0;
848
849 if (!skb_has_frag_list(skb)) {
850 /*
851 * Only one fragment on the socket.
852 */
853 skb->csum_start = skb_transport_header(skb) - skb->head;
854 skb->csum_offset = offsetof(struct udphdr, check);
855 uh->check = ~csum_tcpudp_magic(src, dst, len,
856 IPPROTO_UDP, 0);
857 } else {
858 struct sk_buff *frags;
859
860 /*
861 * HW-checksum won't work as there are two or more
862 * fragments on the socket so that all csums of sk_buffs
863 * should be together
864 */
865 skb_walk_frags(skb, frags) {
866 csum = csum_add(csum, frags->csum);
867 hlen -= frags->len;
868 }
869
870 csum = skb_checksum(skb, offset, hlen, csum);
871 skb->ip_summed = CHECKSUM_NONE;
872
873 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
874 if (uh->check == 0)
875 uh->check = CSUM_MANGLED_0;
876 }
877 }
878 EXPORT_SYMBOL_GPL(udp4_hwcsum);
879
880 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
881 * for the simple case like when setting the checksum for a UDP tunnel.
882 */
udp_set_csum(bool nocheck,struct sk_buff * skb,__be32 saddr,__be32 daddr,int len)883 void udp_set_csum(bool nocheck, struct sk_buff *skb,
884 __be32 saddr, __be32 daddr, int len)
885 {
886 struct udphdr *uh = udp_hdr(skb);
887
888 if (nocheck) {
889 uh->check = 0;
890 } else if (skb_is_gso(skb)) {
891 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
892 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
893 uh->check = 0;
894 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
895 if (uh->check == 0)
896 uh->check = CSUM_MANGLED_0;
897 } else {
898 skb->ip_summed = CHECKSUM_PARTIAL;
899 skb->csum_start = skb_transport_header(skb) - skb->head;
900 skb->csum_offset = offsetof(struct udphdr, check);
901 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
902 }
903 }
904 EXPORT_SYMBOL(udp_set_csum);
905
udp_send_skb(struct sk_buff * skb,struct flowi4 * fl4,struct inet_cork * cork)906 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
907 struct inet_cork *cork)
908 {
909 struct sock *sk = skb->sk;
910 struct inet_sock *inet = inet_sk(sk);
911 struct udphdr *uh;
912 int err;
913 int is_udplite = IS_UDPLITE(sk);
914 int offset = skb_transport_offset(skb);
915 int len = skb->len - offset;
916 int datalen = len - sizeof(*uh);
917 __wsum csum = 0;
918
919 /*
920 * Create a UDP header
921 */
922 uh = udp_hdr(skb);
923 uh->source = inet->inet_sport;
924 uh->dest = fl4->fl4_dport;
925 uh->len = htons(len);
926 uh->check = 0;
927
928 if (cork->gso_size) {
929 const int hlen = skb_network_header_len(skb) +
930 sizeof(struct udphdr);
931
932 if (hlen + cork->gso_size > cork->fragsize) {
933 kfree_skb(skb);
934 return -EINVAL;
935 }
936 if (datalen > cork->gso_size * UDP_MAX_SEGMENTS) {
937 kfree_skb(skb);
938 return -EINVAL;
939 }
940 if (sk->sk_no_check_tx) {
941 kfree_skb(skb);
942 return -EINVAL;
943 }
944 if (is_udplite || dst_xfrm(skb_dst(skb))) {
945 kfree_skb(skb);
946 return -EIO;
947 }
948
949 if (datalen > cork->gso_size) {
950 skb_shinfo(skb)->gso_size = cork->gso_size;
951 skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
952 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
953 cork->gso_size);
954 }
955 goto csum_partial;
956 }
957
958 if (is_udplite) /* UDP-Lite */
959 csum = udplite_csum(skb);
960
961 else if (sk->sk_no_check_tx) { /* UDP csum off */
962
963 skb->ip_summed = CHECKSUM_NONE;
964 goto send;
965
966 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
967 csum_partial:
968
969 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
970 goto send;
971
972 } else
973 csum = udp_csum(skb);
974
975 /* add protocol-dependent pseudo-header */
976 uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
977 sk->sk_protocol, csum);
978 if (uh->check == 0)
979 uh->check = CSUM_MANGLED_0;
980
981 send:
982 err = ip_send_skb(sock_net(sk), skb);
983 if (err) {
984 if (err == -ENOBUFS &&
985 !inet_test_bit(RECVERR, sk)) {
986 UDP_INC_STATS(sock_net(sk),
987 UDP_MIB_SNDBUFERRORS, is_udplite);
988 err = 0;
989 }
990 } else
991 UDP_INC_STATS(sock_net(sk),
992 UDP_MIB_OUTDATAGRAMS, is_udplite);
993 return err;
994 }
995
996 /*
997 * Push out all pending data as one UDP datagram. Socket is locked.
998 */
udp_push_pending_frames(struct sock * sk)999 int udp_push_pending_frames(struct sock *sk)
1000 {
1001 struct udp_sock *up = udp_sk(sk);
1002 struct inet_sock *inet = inet_sk(sk);
1003 struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
1004 struct sk_buff *skb;
1005 int err = 0;
1006
1007 skb = ip_finish_skb(sk, fl4);
1008 if (!skb)
1009 goto out;
1010
1011 err = udp_send_skb(skb, fl4, &inet->cork.base);
1012
1013 out:
1014 up->len = 0;
1015 WRITE_ONCE(up->pending, 0);
1016 return err;
1017 }
1018 EXPORT_SYMBOL(udp_push_pending_frames);
1019
__udp_cmsg_send(struct cmsghdr * cmsg,u16 * gso_size)1020 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
1021 {
1022 switch (cmsg->cmsg_type) {
1023 case UDP_SEGMENT:
1024 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
1025 return -EINVAL;
1026 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
1027 return 0;
1028 default:
1029 return -EINVAL;
1030 }
1031 }
1032
udp_cmsg_send(struct sock * sk,struct msghdr * msg,u16 * gso_size)1033 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
1034 {
1035 struct cmsghdr *cmsg;
1036 bool need_ip = false;
1037 int err;
1038
1039 for_each_cmsghdr(cmsg, msg) {
1040 if (!CMSG_OK(msg, cmsg))
1041 return -EINVAL;
1042
1043 if (cmsg->cmsg_level != SOL_UDP) {
1044 need_ip = true;
1045 continue;
1046 }
1047
1048 err = __udp_cmsg_send(cmsg, gso_size);
1049 if (err)
1050 return err;
1051 }
1052
1053 return need_ip;
1054 }
1055 EXPORT_SYMBOL_GPL(udp_cmsg_send);
1056
udp_sendmsg(struct sock * sk,struct msghdr * msg,size_t len)1057 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
1058 {
1059 struct inet_sock *inet = inet_sk(sk);
1060 struct udp_sock *up = udp_sk(sk);
1061 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
1062 struct flowi4 fl4_stack;
1063 struct flowi4 *fl4;
1064 int ulen = len;
1065 struct ipcm_cookie ipc;
1066 struct rtable *rt = NULL;
1067 int free = 0;
1068 int connected = 0;
1069 __be32 daddr, faddr, saddr;
1070 u8 tos, scope;
1071 __be16 dport;
1072 int err, is_udplite = IS_UDPLITE(sk);
1073 int corkreq = udp_test_bit(CORK, sk) || msg->msg_flags & MSG_MORE;
1074 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
1075 struct sk_buff *skb;
1076 struct ip_options_data opt_copy;
1077 int uc_index;
1078
1079 if (len > 0xFFFF)
1080 return -EMSGSIZE;
1081
1082 /*
1083 * Check the flags.
1084 */
1085
1086 if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
1087 return -EOPNOTSUPP;
1088
1089 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
1090
1091 fl4 = &inet->cork.fl.u.ip4;
1092 if (READ_ONCE(up->pending)) {
1093 /*
1094 * There are pending frames.
1095 * The socket lock must be held while it's corked.
1096 */
1097 lock_sock(sk);
1098 if (likely(up->pending)) {
1099 if (unlikely(up->pending != AF_INET)) {
1100 release_sock(sk);
1101 return -EINVAL;
1102 }
1103 goto do_append_data;
1104 }
1105 release_sock(sk);
1106 }
1107 ulen += sizeof(struct udphdr);
1108
1109 /*
1110 * Get and verify the address.
1111 */
1112 if (usin) {
1113 if (msg->msg_namelen < sizeof(*usin))
1114 return -EINVAL;
1115 if (usin->sin_family != AF_INET) {
1116 if (usin->sin_family != AF_UNSPEC)
1117 return -EAFNOSUPPORT;
1118 }
1119
1120 daddr = usin->sin_addr.s_addr;
1121 dport = usin->sin_port;
1122 if (dport == 0)
1123 return -EINVAL;
1124 } else {
1125 if (sk->sk_state != TCP_ESTABLISHED)
1126 return -EDESTADDRREQ;
1127 daddr = inet->inet_daddr;
1128 dport = inet->inet_dport;
1129 /* Open fast path for connected socket.
1130 Route will not be used, if at least one option is set.
1131 */
1132 connected = 1;
1133 }
1134
1135 ipcm_init_sk(&ipc, inet);
1136 ipc.gso_size = READ_ONCE(up->gso_size);
1137
1138 if (msg->msg_controllen) {
1139 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1140 if (err > 0) {
1141 err = ip_cmsg_send(sk, msg, &ipc,
1142 sk->sk_family == AF_INET6);
1143 connected = 0;
1144 }
1145 if (unlikely(err < 0)) {
1146 kfree(ipc.opt);
1147 return err;
1148 }
1149 if (ipc.opt)
1150 free = 1;
1151 }
1152 if (!ipc.opt) {
1153 struct ip_options_rcu *inet_opt;
1154
1155 rcu_read_lock();
1156 inet_opt = rcu_dereference(inet->inet_opt);
1157 if (inet_opt) {
1158 memcpy(&opt_copy, inet_opt,
1159 sizeof(*inet_opt) + inet_opt->opt.optlen);
1160 ipc.opt = &opt_copy.opt;
1161 }
1162 rcu_read_unlock();
1163 }
1164
1165 if (cgroup_bpf_enabled(CGROUP_UDP4_SENDMSG) && !connected) {
1166 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1167 (struct sockaddr *)usin,
1168 &msg->msg_namelen,
1169 &ipc.addr);
1170 if (err)
1171 goto out_free;
1172 if (usin) {
1173 if (usin->sin_port == 0) {
1174 /* BPF program set invalid port. Reject it. */
1175 err = -EINVAL;
1176 goto out_free;
1177 }
1178 daddr = usin->sin_addr.s_addr;
1179 dport = usin->sin_port;
1180 }
1181 }
1182
1183 saddr = ipc.addr;
1184 ipc.addr = faddr = daddr;
1185
1186 if (ipc.opt && ipc.opt->opt.srr) {
1187 if (!daddr) {
1188 err = -EINVAL;
1189 goto out_free;
1190 }
1191 faddr = ipc.opt->opt.faddr;
1192 connected = 0;
1193 }
1194 tos = get_rttos(&ipc, inet);
1195 scope = ip_sendmsg_scope(inet, &ipc, msg);
1196 if (scope == RT_SCOPE_LINK)
1197 connected = 0;
1198
1199 uc_index = READ_ONCE(inet->uc_index);
1200 if (ipv4_is_multicast(daddr)) {
1201 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1202 ipc.oif = READ_ONCE(inet->mc_index);
1203 if (!saddr)
1204 saddr = READ_ONCE(inet->mc_addr);
1205 connected = 0;
1206 } else if (!ipc.oif) {
1207 ipc.oif = uc_index;
1208 } else if (ipv4_is_lbcast(daddr) && uc_index) {
1209 /* oif is set, packet is to local broadcast and
1210 * uc_index is set. oif is most likely set
1211 * by sk_bound_dev_if. If uc_index != oif check if the
1212 * oif is an L3 master and uc_index is an L3 slave.
1213 * If so, we want to allow the send using the uc_index.
1214 */
1215 if (ipc.oif != uc_index &&
1216 ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1217 uc_index)) {
1218 ipc.oif = uc_index;
1219 }
1220 }
1221
1222 if (connected)
1223 rt = dst_rtable(sk_dst_check(sk, 0));
1224
1225 if (!rt) {
1226 struct net *net = sock_net(sk);
1227 __u8 flow_flags = inet_sk_flowi_flags(sk);
1228
1229 fl4 = &fl4_stack;
1230
1231 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos, scope,
1232 sk->sk_protocol, flow_flags, faddr, saddr,
1233 dport, inet->inet_sport, sk->sk_uid);
1234
1235 security_sk_classify_flow(sk, flowi4_to_flowi_common(fl4));
1236 rt = ip_route_output_flow(net, fl4, sk);
1237 if (IS_ERR(rt)) {
1238 err = PTR_ERR(rt);
1239 rt = NULL;
1240 if (err == -ENETUNREACH)
1241 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1242 goto out;
1243 }
1244
1245 err = -EACCES;
1246 if ((rt->rt_flags & RTCF_BROADCAST) &&
1247 !sock_flag(sk, SOCK_BROADCAST))
1248 goto out;
1249 if (connected)
1250 sk_dst_set(sk, dst_clone(&rt->dst));
1251 }
1252
1253 if (msg->msg_flags&MSG_CONFIRM)
1254 goto do_confirm;
1255 back_from_confirm:
1256
1257 saddr = fl4->saddr;
1258 if (!ipc.addr)
1259 daddr = ipc.addr = fl4->daddr;
1260
1261 /* Lockless fast path for the non-corking case. */
1262 if (!corkreq) {
1263 struct inet_cork cork;
1264
1265 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1266 sizeof(struct udphdr), &ipc, &rt,
1267 &cork, msg->msg_flags);
1268 err = PTR_ERR(skb);
1269 if (!IS_ERR_OR_NULL(skb))
1270 err = udp_send_skb(skb, fl4, &cork);
1271 goto out;
1272 }
1273
1274 lock_sock(sk);
1275 if (unlikely(up->pending)) {
1276 /* The socket is already corked while preparing it. */
1277 /* ... which is an evident application bug. --ANK */
1278 release_sock(sk);
1279
1280 net_dbg_ratelimited("socket already corked\n");
1281 err = -EINVAL;
1282 goto out;
1283 }
1284 /*
1285 * Now cork the socket to pend data.
1286 */
1287 fl4 = &inet->cork.fl.u.ip4;
1288 fl4->daddr = daddr;
1289 fl4->saddr = saddr;
1290 fl4->fl4_dport = dport;
1291 fl4->fl4_sport = inet->inet_sport;
1292 WRITE_ONCE(up->pending, AF_INET);
1293
1294 do_append_data:
1295 up->len += ulen;
1296 err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1297 sizeof(struct udphdr), &ipc, &rt,
1298 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1299 if (err)
1300 udp_flush_pending_frames(sk);
1301 else if (!corkreq)
1302 err = udp_push_pending_frames(sk);
1303 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1304 WRITE_ONCE(up->pending, 0);
1305 release_sock(sk);
1306
1307 out:
1308 ip_rt_put(rt);
1309 out_free:
1310 if (free)
1311 kfree(ipc.opt);
1312 if (!err)
1313 return len;
1314 /*
1315 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
1316 * ENOBUFS might not be good (it's not tunable per se), but otherwise
1317 * we don't have a good statistic (IpOutDiscards but it can be too many
1318 * things). We could add another new stat but at least for now that
1319 * seems like overkill.
1320 */
1321 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1322 UDP_INC_STATS(sock_net(sk),
1323 UDP_MIB_SNDBUFERRORS, is_udplite);
1324 }
1325 return err;
1326
1327 do_confirm:
1328 if (msg->msg_flags & MSG_PROBE)
1329 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1330 if (!(msg->msg_flags&MSG_PROBE) || len)
1331 goto back_from_confirm;
1332 err = 0;
1333 goto out;
1334 }
1335 EXPORT_SYMBOL(udp_sendmsg);
1336
udp_splice_eof(struct socket * sock)1337 void udp_splice_eof(struct socket *sock)
1338 {
1339 struct sock *sk = sock->sk;
1340 struct udp_sock *up = udp_sk(sk);
1341
1342 if (!READ_ONCE(up->pending) || udp_test_bit(CORK, sk))
1343 return;
1344
1345 lock_sock(sk);
1346 if (up->pending && !udp_test_bit(CORK, sk))
1347 udp_push_pending_frames(sk);
1348 release_sock(sk);
1349 }
1350 EXPORT_SYMBOL_GPL(udp_splice_eof);
1351
1352 #define UDP_SKB_IS_STATELESS 0x80000000
1353
1354 /* all head states (dst, sk, nf conntrack) except skb extensions are
1355 * cleared by udp_rcv().
1356 *
1357 * We need to preserve secpath, if present, to eventually process
1358 * IP_CMSG_PASSSEC at recvmsg() time.
1359 *
1360 * Other extensions can be cleared.
1361 */
udp_try_make_stateless(struct sk_buff * skb)1362 static bool udp_try_make_stateless(struct sk_buff *skb)
1363 {
1364 if (!skb_has_extensions(skb))
1365 return true;
1366
1367 if (!secpath_exists(skb)) {
1368 skb_ext_reset(skb);
1369 return true;
1370 }
1371
1372 return false;
1373 }
1374
udp_set_dev_scratch(struct sk_buff * skb)1375 static void udp_set_dev_scratch(struct sk_buff *skb)
1376 {
1377 struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1378
1379 BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1380 scratch->_tsize_state = skb->truesize;
1381 #if BITS_PER_LONG == 64
1382 scratch->len = skb->len;
1383 scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1384 scratch->is_linear = !skb_is_nonlinear(skb);
1385 #endif
1386 if (udp_try_make_stateless(skb))
1387 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1388 }
1389
udp_skb_csum_unnecessary_set(struct sk_buff * skb)1390 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1391 {
1392 /* We come here after udp_lib_checksum_complete() returned 0.
1393 * This means that __skb_checksum_complete() might have
1394 * set skb->csum_valid to 1.
1395 * On 64bit platforms, we can set csum_unnecessary
1396 * to true, but only if the skb is not shared.
1397 */
1398 #if BITS_PER_LONG == 64
1399 if (!skb_shared(skb))
1400 udp_skb_scratch(skb)->csum_unnecessary = true;
1401 #endif
1402 }
1403
udp_skb_truesize(struct sk_buff * skb)1404 static int udp_skb_truesize(struct sk_buff *skb)
1405 {
1406 return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1407 }
1408
udp_skb_has_head_state(struct sk_buff * skb)1409 static bool udp_skb_has_head_state(struct sk_buff *skb)
1410 {
1411 return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1412 }
1413
1414 /* fully reclaim rmem/fwd memory allocated for skb */
udp_rmem_release(struct sock * sk,int size,int partial,bool rx_queue_lock_held)1415 static void udp_rmem_release(struct sock *sk, int size, int partial,
1416 bool rx_queue_lock_held)
1417 {
1418 struct udp_sock *up = udp_sk(sk);
1419 struct sk_buff_head *sk_queue;
1420 int amt;
1421
1422 if (likely(partial)) {
1423 up->forward_deficit += size;
1424 size = up->forward_deficit;
1425 if (size < READ_ONCE(up->forward_threshold) &&
1426 !skb_queue_empty(&up->reader_queue))
1427 return;
1428 } else {
1429 size += up->forward_deficit;
1430 }
1431 up->forward_deficit = 0;
1432
1433 /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1434 * if the called don't held it already
1435 */
1436 sk_queue = &sk->sk_receive_queue;
1437 if (!rx_queue_lock_held)
1438 spin_lock(&sk_queue->lock);
1439
1440
1441 sk_forward_alloc_add(sk, size);
1442 amt = (sk->sk_forward_alloc - partial) & ~(PAGE_SIZE - 1);
1443 sk_forward_alloc_add(sk, -amt);
1444
1445 if (amt)
1446 __sk_mem_reduce_allocated(sk, amt >> PAGE_SHIFT);
1447
1448 atomic_sub(size, &sk->sk_rmem_alloc);
1449
1450 /* this can save us from acquiring the rx queue lock on next receive */
1451 skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1452
1453 if (!rx_queue_lock_held)
1454 spin_unlock(&sk_queue->lock);
1455 }
1456
1457 /* Note: called with reader_queue.lock held.
1458 * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1459 * This avoids a cache line miss while receive_queue lock is held.
1460 * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1461 */
udp_skb_destructor(struct sock * sk,struct sk_buff * skb)1462 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1463 {
1464 prefetch(&skb->data);
1465 udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1466 }
1467 EXPORT_SYMBOL(udp_skb_destructor);
1468
1469 /* as above, but the caller held the rx queue lock, too */
udp_skb_dtor_locked(struct sock * sk,struct sk_buff * skb)1470 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1471 {
1472 prefetch(&skb->data);
1473 udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1474 }
1475
1476 /* Idea of busylocks is to let producers grab an extra spinlock
1477 * to relieve pressure on the receive_queue spinlock shared by consumer.
1478 * Under flood, this means that only one producer can be in line
1479 * trying to acquire the receive_queue spinlock.
1480 * These busylock can be allocated on a per cpu manner, instead of a
1481 * per socket one (that would consume a cache line per socket)
1482 */
1483 static int udp_busylocks_log __read_mostly;
1484 static spinlock_t *udp_busylocks __read_mostly;
1485
busylock_acquire(void * ptr)1486 static spinlock_t *busylock_acquire(void *ptr)
1487 {
1488 spinlock_t *busy;
1489
1490 busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1491 spin_lock(busy);
1492 return busy;
1493 }
1494
busylock_release(spinlock_t * busy)1495 static void busylock_release(spinlock_t *busy)
1496 {
1497 if (busy)
1498 spin_unlock(busy);
1499 }
1500
udp_rmem_schedule(struct sock * sk,int size)1501 static int udp_rmem_schedule(struct sock *sk, int size)
1502 {
1503 int delta;
1504
1505 delta = size - sk->sk_forward_alloc;
1506 if (delta > 0 && !__sk_mem_schedule(sk, delta, SK_MEM_RECV))
1507 return -ENOBUFS;
1508
1509 return 0;
1510 }
1511
__udp_enqueue_schedule_skb(struct sock * sk,struct sk_buff * skb)1512 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1513 {
1514 struct sk_buff_head *list = &sk->sk_receive_queue;
1515 int rmem, err = -ENOMEM;
1516 spinlock_t *busy = NULL;
1517 bool becomes_readable;
1518 int size, rcvbuf;
1519
1520 /* Immediately drop when the receive queue is full.
1521 * Always allow at least one packet.
1522 */
1523 rmem = atomic_read(&sk->sk_rmem_alloc);
1524 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1525 if (rmem > rcvbuf)
1526 goto drop;
1527
1528 /* Under mem pressure, it might be helpful to help udp_recvmsg()
1529 * having linear skbs :
1530 * - Reduce memory overhead and thus increase receive queue capacity
1531 * - Less cache line misses at copyout() time
1532 * - Less work at consume_skb() (less alien page frag freeing)
1533 */
1534 if (rmem > (rcvbuf >> 1)) {
1535 skb_condense(skb);
1536
1537 busy = busylock_acquire(sk);
1538 }
1539 size = skb->truesize;
1540 udp_set_dev_scratch(skb);
1541
1542 atomic_add(size, &sk->sk_rmem_alloc);
1543
1544 spin_lock(&list->lock);
1545 err = udp_rmem_schedule(sk, size);
1546 if (err) {
1547 spin_unlock(&list->lock);
1548 goto uncharge_drop;
1549 }
1550
1551 sk_forward_alloc_add(sk, -size);
1552
1553 /* no need to setup a destructor, we will explicitly release the
1554 * forward allocated memory on dequeue
1555 */
1556 sock_skb_set_dropcount(sk, skb);
1557
1558 becomes_readable = skb_queue_empty(list);
1559 __skb_queue_tail(list, skb);
1560 spin_unlock(&list->lock);
1561
1562 if (!sock_flag(sk, SOCK_DEAD)) {
1563 if (becomes_readable ||
1564 sk->sk_data_ready != sock_def_readable ||
1565 READ_ONCE(sk->sk_peek_off) >= 0)
1566 INDIRECT_CALL_1(sk->sk_data_ready,
1567 sock_def_readable, sk);
1568 else
1569 sk_wake_async_rcu(sk, SOCK_WAKE_WAITD, POLL_IN);
1570 }
1571 busylock_release(busy);
1572 return 0;
1573
1574 uncharge_drop:
1575 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1576
1577 drop:
1578 atomic_inc(&sk->sk_drops);
1579 busylock_release(busy);
1580 return err;
1581 }
1582 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1583
udp_destruct_common(struct sock * sk)1584 void udp_destruct_common(struct sock *sk)
1585 {
1586 /* reclaim completely the forward allocated memory */
1587 struct udp_sock *up = udp_sk(sk);
1588 unsigned int total = 0;
1589 struct sk_buff *skb;
1590
1591 skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1592 while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1593 total += skb->truesize;
1594 kfree_skb(skb);
1595 }
1596 udp_rmem_release(sk, total, 0, true);
1597 }
1598 EXPORT_SYMBOL_GPL(udp_destruct_common);
1599
udp_destruct_sock(struct sock * sk)1600 static void udp_destruct_sock(struct sock *sk)
1601 {
1602 udp_destruct_common(sk);
1603 inet_sock_destruct(sk);
1604 }
1605
udp_init_sock(struct sock * sk)1606 int udp_init_sock(struct sock *sk)
1607 {
1608 udp_lib_init_sock(sk);
1609 sk->sk_destruct = udp_destruct_sock;
1610 set_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1611 return 0;
1612 }
1613
skb_consume_udp(struct sock * sk,struct sk_buff * skb,int len)1614 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1615 {
1616 if (unlikely(READ_ONCE(udp_sk(sk)->peeking_with_offset)))
1617 sk_peek_offset_bwd(sk, len);
1618
1619 if (!skb_unref(skb))
1620 return;
1621
1622 /* In the more common cases we cleared the head states previously,
1623 * see __udp_queue_rcv_skb().
1624 */
1625 if (unlikely(udp_skb_has_head_state(skb)))
1626 skb_release_head_state(skb);
1627 __consume_stateless_skb(skb);
1628 }
1629 EXPORT_SYMBOL_GPL(skb_consume_udp);
1630
__first_packet_length(struct sock * sk,struct sk_buff_head * rcvq,int * total)1631 static struct sk_buff *__first_packet_length(struct sock *sk,
1632 struct sk_buff_head *rcvq,
1633 int *total)
1634 {
1635 struct sk_buff *skb;
1636
1637 while ((skb = skb_peek(rcvq)) != NULL) {
1638 if (udp_lib_checksum_complete(skb)) {
1639 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1640 IS_UDPLITE(sk));
1641 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1642 IS_UDPLITE(sk));
1643 atomic_inc(&sk->sk_drops);
1644 __skb_unlink(skb, rcvq);
1645 *total += skb->truesize;
1646 kfree_skb(skb);
1647 } else {
1648 udp_skb_csum_unnecessary_set(skb);
1649 break;
1650 }
1651 }
1652 return skb;
1653 }
1654
1655 /**
1656 * first_packet_length - return length of first packet in receive queue
1657 * @sk: socket
1658 *
1659 * Drops all bad checksum frames, until a valid one is found.
1660 * Returns the length of found skb, or -1 if none is found.
1661 */
first_packet_length(struct sock * sk)1662 static int first_packet_length(struct sock *sk)
1663 {
1664 struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1665 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1666 struct sk_buff *skb;
1667 int total = 0;
1668 int res;
1669
1670 spin_lock_bh(&rcvq->lock);
1671 skb = __first_packet_length(sk, rcvq, &total);
1672 if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1673 spin_lock(&sk_queue->lock);
1674 skb_queue_splice_tail_init(sk_queue, rcvq);
1675 spin_unlock(&sk_queue->lock);
1676
1677 skb = __first_packet_length(sk, rcvq, &total);
1678 }
1679 res = skb ? skb->len : -1;
1680 if (total)
1681 udp_rmem_release(sk, total, 1, false);
1682 spin_unlock_bh(&rcvq->lock);
1683 return res;
1684 }
1685
1686 /*
1687 * IOCTL requests applicable to the UDP protocol
1688 */
1689
udp_ioctl(struct sock * sk,int cmd,int * karg)1690 int udp_ioctl(struct sock *sk, int cmd, int *karg)
1691 {
1692 switch (cmd) {
1693 case SIOCOUTQ:
1694 {
1695 *karg = sk_wmem_alloc_get(sk);
1696 return 0;
1697 }
1698
1699 case SIOCINQ:
1700 {
1701 *karg = max_t(int, 0, first_packet_length(sk));
1702 return 0;
1703 }
1704
1705 default:
1706 return -ENOIOCTLCMD;
1707 }
1708
1709 return 0;
1710 }
1711 EXPORT_SYMBOL(udp_ioctl);
1712
__skb_recv_udp(struct sock * sk,unsigned int flags,int * off,int * err)1713 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1714 int *off, int *err)
1715 {
1716 struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1717 struct sk_buff_head *queue;
1718 struct sk_buff *last;
1719 long timeo;
1720 int error;
1721
1722 queue = &udp_sk(sk)->reader_queue;
1723 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1724 do {
1725 struct sk_buff *skb;
1726
1727 error = sock_error(sk);
1728 if (error)
1729 break;
1730
1731 error = -EAGAIN;
1732 do {
1733 spin_lock_bh(&queue->lock);
1734 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1735 err, &last);
1736 if (skb) {
1737 if (!(flags & MSG_PEEK))
1738 udp_skb_destructor(sk, skb);
1739 spin_unlock_bh(&queue->lock);
1740 return skb;
1741 }
1742
1743 if (skb_queue_empty_lockless(sk_queue)) {
1744 spin_unlock_bh(&queue->lock);
1745 goto busy_check;
1746 }
1747
1748 /* refill the reader queue and walk it again
1749 * keep both queues locked to avoid re-acquiring
1750 * the sk_receive_queue lock if fwd memory scheduling
1751 * is needed.
1752 */
1753 spin_lock(&sk_queue->lock);
1754 skb_queue_splice_tail_init(sk_queue, queue);
1755
1756 skb = __skb_try_recv_from_queue(sk, queue, flags, off,
1757 err, &last);
1758 if (skb && !(flags & MSG_PEEK))
1759 udp_skb_dtor_locked(sk, skb);
1760 spin_unlock(&sk_queue->lock);
1761 spin_unlock_bh(&queue->lock);
1762 if (skb)
1763 return skb;
1764
1765 busy_check:
1766 if (!sk_can_busy_loop(sk))
1767 break;
1768
1769 sk_busy_loop(sk, flags & MSG_DONTWAIT);
1770 } while (!skb_queue_empty_lockless(sk_queue));
1771
1772 /* sk_queue is empty, reader_queue may contain peeked packets */
1773 } while (timeo &&
1774 !__skb_wait_for_more_packets(sk, &sk->sk_receive_queue,
1775 &error, &timeo,
1776 (struct sk_buff *)sk_queue));
1777
1778 *err = error;
1779 return NULL;
1780 }
1781 EXPORT_SYMBOL(__skb_recv_udp);
1782
udp_read_skb(struct sock * sk,skb_read_actor_t recv_actor)1783 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor)
1784 {
1785 struct sk_buff *skb;
1786 int err;
1787
1788 try_again:
1789 skb = skb_recv_udp(sk, MSG_DONTWAIT, &err);
1790 if (!skb)
1791 return err;
1792
1793 if (udp_lib_checksum_complete(skb)) {
1794 int is_udplite = IS_UDPLITE(sk);
1795 struct net *net = sock_net(sk);
1796
1797 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, is_udplite);
1798 __UDP_INC_STATS(net, UDP_MIB_INERRORS, is_udplite);
1799 atomic_inc(&sk->sk_drops);
1800 kfree_skb(skb);
1801 goto try_again;
1802 }
1803
1804 WARN_ON_ONCE(!skb_set_owner_sk_safe(skb, sk));
1805 return recv_actor(sk, skb);
1806 }
1807 EXPORT_SYMBOL(udp_read_skb);
1808
1809 /*
1810 * This should be easy, if there is something there we
1811 * return it, otherwise we block.
1812 */
1813
udp_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)1814 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int flags,
1815 int *addr_len)
1816 {
1817 struct inet_sock *inet = inet_sk(sk);
1818 DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1819 struct sk_buff *skb;
1820 unsigned int ulen, copied;
1821 int off, err, peeking = flags & MSG_PEEK;
1822 int is_udplite = IS_UDPLITE(sk);
1823 bool checksum_valid = false;
1824
1825 if (flags & MSG_ERRQUEUE)
1826 return ip_recv_error(sk, msg, len, addr_len);
1827
1828 try_again:
1829 off = sk_peek_offset(sk, flags);
1830 skb = __skb_recv_udp(sk, flags, &off, &err);
1831 if (!skb)
1832 return err;
1833
1834 ulen = udp_skb_len(skb);
1835 copied = len;
1836 if (copied > ulen - off)
1837 copied = ulen - off;
1838 else if (copied < ulen)
1839 msg->msg_flags |= MSG_TRUNC;
1840
1841 /*
1842 * If checksum is needed at all, try to do it while copying the
1843 * data. If the data is truncated, or if we only want a partial
1844 * coverage checksum (UDP-Lite), do it before the copy.
1845 */
1846
1847 if (copied < ulen || peeking ||
1848 (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1849 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1850 !__udp_lib_checksum_complete(skb);
1851 if (!checksum_valid)
1852 goto csum_copy_err;
1853 }
1854
1855 if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1856 if (udp_skb_is_linear(skb))
1857 err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1858 else
1859 err = skb_copy_datagram_msg(skb, off, msg, copied);
1860 } else {
1861 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1862
1863 if (err == -EINVAL)
1864 goto csum_copy_err;
1865 }
1866
1867 if (unlikely(err)) {
1868 if (!peeking) {
1869 atomic_inc(&sk->sk_drops);
1870 UDP_INC_STATS(sock_net(sk),
1871 UDP_MIB_INERRORS, is_udplite);
1872 }
1873 kfree_skb(skb);
1874 return err;
1875 }
1876
1877 if (!peeking)
1878 UDP_INC_STATS(sock_net(sk),
1879 UDP_MIB_INDATAGRAMS, is_udplite);
1880
1881 sock_recv_cmsgs(msg, sk, skb);
1882
1883 /* Copy the address. */
1884 if (sin) {
1885 sin->sin_family = AF_INET;
1886 sin->sin_port = udp_hdr(skb)->source;
1887 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1888 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1889 *addr_len = sizeof(*sin);
1890
1891 BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1892 (struct sockaddr *)sin,
1893 addr_len);
1894 }
1895
1896 if (udp_test_bit(GRO_ENABLED, sk))
1897 udp_cmsg_recv(msg, sk, skb);
1898
1899 if (inet_cmsg_flags(inet))
1900 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1901
1902 err = copied;
1903 if (flags & MSG_TRUNC)
1904 err = ulen;
1905
1906 skb_consume_udp(sk, skb, peeking ? -err : err);
1907 return err;
1908
1909 csum_copy_err:
1910 if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1911 udp_skb_destructor)) {
1912 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1913 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1914 }
1915 kfree_skb(skb);
1916
1917 /* starting over for a new packet, but check if we need to yield */
1918 cond_resched();
1919 msg->msg_flags &= ~MSG_TRUNC;
1920 goto try_again;
1921 }
1922
udp_pre_connect(struct sock * sk,struct sockaddr * uaddr,int addr_len)1923 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1924 {
1925 /* This check is replicated from __ip4_datagram_connect() and
1926 * intended to prevent BPF program called below from accessing bytes
1927 * that are out of the bound specified by user in addr_len.
1928 */
1929 if (addr_len < sizeof(struct sockaddr_in))
1930 return -EINVAL;
1931
1932 return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr, &addr_len);
1933 }
1934 EXPORT_SYMBOL(udp_pre_connect);
1935
__udp_disconnect(struct sock * sk,int flags)1936 int __udp_disconnect(struct sock *sk, int flags)
1937 {
1938 struct inet_sock *inet = inet_sk(sk);
1939 /*
1940 * 1003.1g - break association.
1941 */
1942
1943 sk->sk_state = TCP_CLOSE;
1944 inet->inet_daddr = 0;
1945 inet->inet_dport = 0;
1946 sock_rps_reset_rxhash(sk);
1947 sk->sk_bound_dev_if = 0;
1948 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1949 inet_reset_saddr(sk);
1950 if (sk->sk_prot->rehash &&
1951 (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1952 sk->sk_prot->rehash(sk);
1953 }
1954
1955 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1956 sk->sk_prot->unhash(sk);
1957 inet->inet_sport = 0;
1958 }
1959 sk_dst_reset(sk);
1960 return 0;
1961 }
1962 EXPORT_SYMBOL(__udp_disconnect);
1963
udp_disconnect(struct sock * sk,int flags)1964 int udp_disconnect(struct sock *sk, int flags)
1965 {
1966 lock_sock(sk);
1967 __udp_disconnect(sk, flags);
1968 release_sock(sk);
1969 return 0;
1970 }
1971 EXPORT_SYMBOL(udp_disconnect);
1972
udp_lib_unhash(struct sock * sk)1973 void udp_lib_unhash(struct sock *sk)
1974 {
1975 if (sk_hashed(sk)) {
1976 struct udp_table *udptable = udp_get_table_prot(sk);
1977 struct udp_hslot *hslot, *hslot2;
1978
1979 hslot = udp_hashslot(udptable, sock_net(sk),
1980 udp_sk(sk)->udp_port_hash);
1981 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1982
1983 spin_lock_bh(&hslot->lock);
1984 if (rcu_access_pointer(sk->sk_reuseport_cb))
1985 reuseport_detach_sock(sk);
1986 if (sk_del_node_init_rcu(sk)) {
1987 hslot->count--;
1988 inet_sk(sk)->inet_num = 0;
1989 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1990
1991 spin_lock(&hslot2->lock);
1992 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1993 hslot2->count--;
1994 spin_unlock(&hslot2->lock);
1995 }
1996 spin_unlock_bh(&hslot->lock);
1997 }
1998 }
1999 EXPORT_SYMBOL(udp_lib_unhash);
2000
2001 /*
2002 * inet_rcv_saddr was changed, we must rehash secondary hash
2003 */
udp_lib_rehash(struct sock * sk,u16 newhash)2004 void udp_lib_rehash(struct sock *sk, u16 newhash)
2005 {
2006 if (sk_hashed(sk)) {
2007 struct udp_table *udptable = udp_get_table_prot(sk);
2008 struct udp_hslot *hslot, *hslot2, *nhslot2;
2009
2010 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
2011 nhslot2 = udp_hashslot2(udptable, newhash);
2012 udp_sk(sk)->udp_portaddr_hash = newhash;
2013
2014 if (hslot2 != nhslot2 ||
2015 rcu_access_pointer(sk->sk_reuseport_cb)) {
2016 hslot = udp_hashslot(udptable, sock_net(sk),
2017 udp_sk(sk)->udp_port_hash);
2018 /* we must lock primary chain too */
2019 spin_lock_bh(&hslot->lock);
2020 if (rcu_access_pointer(sk->sk_reuseport_cb))
2021 reuseport_detach_sock(sk);
2022
2023 if (hslot2 != nhslot2) {
2024 spin_lock(&hslot2->lock);
2025 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
2026 hslot2->count--;
2027 spin_unlock(&hslot2->lock);
2028
2029 spin_lock(&nhslot2->lock);
2030 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
2031 &nhslot2->head);
2032 nhslot2->count++;
2033 spin_unlock(&nhslot2->lock);
2034 }
2035
2036 spin_unlock_bh(&hslot->lock);
2037 }
2038 }
2039 }
2040 EXPORT_SYMBOL(udp_lib_rehash);
2041
udp_v4_rehash(struct sock * sk)2042 void udp_v4_rehash(struct sock *sk)
2043 {
2044 u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
2045 inet_sk(sk)->inet_rcv_saddr,
2046 inet_sk(sk)->inet_num);
2047 udp_lib_rehash(sk, new_hash);
2048 }
2049
__udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2050 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2051 {
2052 int rc;
2053
2054 if (inet_sk(sk)->inet_daddr) {
2055 sock_rps_save_rxhash(sk, skb);
2056 sk_mark_napi_id(sk, skb);
2057 sk_incoming_cpu_update(sk);
2058 } else {
2059 sk_mark_napi_id_once(sk, skb);
2060 }
2061
2062 rc = __udp_enqueue_schedule_skb(sk, skb);
2063 if (rc < 0) {
2064 int is_udplite = IS_UDPLITE(sk);
2065 int drop_reason;
2066
2067 /* Note that an ENOMEM error is charged twice */
2068 if (rc == -ENOMEM) {
2069 UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
2070 is_udplite);
2071 drop_reason = SKB_DROP_REASON_SOCKET_RCVBUFF;
2072 } else {
2073 UDP_INC_STATS(sock_net(sk), UDP_MIB_MEMERRORS,
2074 is_udplite);
2075 drop_reason = SKB_DROP_REASON_PROTO_MEM;
2076 }
2077 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2078 trace_udp_fail_queue_rcv_skb(rc, sk, skb);
2079 sk_skb_reason_drop(sk, skb, drop_reason);
2080 return -1;
2081 }
2082
2083 return 0;
2084 }
2085
2086 /* returns:
2087 * -1: error
2088 * 0: success
2089 * >0: "udp encap" protocol resubmission
2090 *
2091 * Note that in the success and error cases, the skb is assumed to
2092 * have either been requeued or freed.
2093 */
udp_queue_rcv_one_skb(struct sock * sk,struct sk_buff * skb)2094 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
2095 {
2096 int drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2097 struct udp_sock *up = udp_sk(sk);
2098 int is_udplite = IS_UDPLITE(sk);
2099
2100 /*
2101 * Charge it to the socket, dropping if the queue is full.
2102 */
2103 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) {
2104 drop_reason = SKB_DROP_REASON_XFRM_POLICY;
2105 goto drop;
2106 }
2107 nf_reset_ct(skb);
2108
2109 if (static_branch_unlikely(&udp_encap_needed_key) &&
2110 READ_ONCE(up->encap_type)) {
2111 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2112
2113 /*
2114 * This is an encapsulation socket so pass the skb to
2115 * the socket's udp_encap_rcv() hook. Otherwise, just
2116 * fall through and pass this up the UDP socket.
2117 * up->encap_rcv() returns the following value:
2118 * =0 if skb was successfully passed to the encap
2119 * handler or was discarded by it.
2120 * >0 if skb should be passed on to UDP.
2121 * <0 if skb should be resubmitted as proto -N
2122 */
2123
2124 /* if we're overly short, let UDP handle it */
2125 encap_rcv = READ_ONCE(up->encap_rcv);
2126 if (encap_rcv) {
2127 int ret;
2128
2129 /* Verify checksum before giving to encap */
2130 if (udp_lib_checksum_complete(skb))
2131 goto csum_error;
2132
2133 ret = encap_rcv(sk, skb);
2134 if (ret <= 0) {
2135 __UDP_INC_STATS(sock_net(sk),
2136 UDP_MIB_INDATAGRAMS,
2137 is_udplite);
2138 return -ret;
2139 }
2140 }
2141
2142 /* FALLTHROUGH -- it's a UDP Packet */
2143 }
2144
2145 /*
2146 * UDP-Lite specific tests, ignored on UDP sockets
2147 */
2148 if (udp_test_bit(UDPLITE_RECV_CC, sk) && UDP_SKB_CB(skb)->partial_cov) {
2149 u16 pcrlen = READ_ONCE(up->pcrlen);
2150
2151 /*
2152 * MIB statistics other than incrementing the error count are
2153 * disabled for the following two types of errors: these depend
2154 * on the application settings, not on the functioning of the
2155 * protocol stack as such.
2156 *
2157 * RFC 3828 here recommends (sec 3.3): "There should also be a
2158 * way ... to ... at least let the receiving application block
2159 * delivery of packets with coverage values less than a value
2160 * provided by the application."
2161 */
2162 if (pcrlen == 0) { /* full coverage was set */
2163 net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2164 UDP_SKB_CB(skb)->cscov, skb->len);
2165 goto drop;
2166 }
2167 /* The next case involves violating the min. coverage requested
2168 * by the receiver. This is subtle: if receiver wants x and x is
2169 * greater than the buffersize/MTU then receiver will complain
2170 * that it wants x while sender emits packets of smaller size y.
2171 * Therefore the above ...()->partial_cov statement is essential.
2172 */
2173 if (UDP_SKB_CB(skb)->cscov < pcrlen) {
2174 net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2175 UDP_SKB_CB(skb)->cscov, pcrlen);
2176 goto drop;
2177 }
2178 }
2179
2180 prefetch(&sk->sk_rmem_alloc);
2181 if (rcu_access_pointer(sk->sk_filter) &&
2182 udp_lib_checksum_complete(skb))
2183 goto csum_error;
2184
2185 if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr))) {
2186 drop_reason = SKB_DROP_REASON_SOCKET_FILTER;
2187 goto drop;
2188 }
2189
2190 udp_csum_pull_header(skb);
2191
2192 ipv4_pktinfo_prepare(sk, skb, true);
2193 return __udp_queue_rcv_skb(sk, skb);
2194
2195 csum_error:
2196 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2197 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2198 drop:
2199 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2200 atomic_inc(&sk->sk_drops);
2201 sk_skb_reason_drop(sk, skb, drop_reason);
2202 return -1;
2203 }
2204
udp_queue_rcv_skb(struct sock * sk,struct sk_buff * skb)2205 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2206 {
2207 struct sk_buff *next, *segs;
2208 int ret;
2209
2210 if (likely(!udp_unexpected_gso(sk, skb)))
2211 return udp_queue_rcv_one_skb(sk, skb);
2212
2213 BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_GSO_CB_OFFSET);
2214 __skb_push(skb, -skb_mac_offset(skb));
2215 segs = udp_rcv_segment(sk, skb, true);
2216 skb_list_walk_safe(segs, skb, next) {
2217 __skb_pull(skb, skb_transport_offset(skb));
2218
2219 udp_post_segment_fix_csum(skb);
2220 ret = udp_queue_rcv_one_skb(sk, skb);
2221 if (ret > 0)
2222 ip_protocol_deliver_rcu(dev_net(skb->dev), skb, ret);
2223 }
2224 return 0;
2225 }
2226
2227 /* For TCP sockets, sk_rx_dst is protected by socket lock
2228 * For UDP, we use xchg() to guard against concurrent changes.
2229 */
udp_sk_rx_dst_set(struct sock * sk,struct dst_entry * dst)2230 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2231 {
2232 struct dst_entry *old;
2233
2234 if (dst_hold_safe(dst)) {
2235 old = unrcu_pointer(xchg(&sk->sk_rx_dst, RCU_INITIALIZER(dst)));
2236 dst_release(old);
2237 return old != dst;
2238 }
2239 return false;
2240 }
2241 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2242
2243 /*
2244 * Multicasts and broadcasts go to each listener.
2245 *
2246 * Note: called only from the BH handler context.
2247 */
__udp4_lib_mcast_deliver(struct net * net,struct sk_buff * skb,struct udphdr * uh,__be32 saddr,__be32 daddr,struct udp_table * udptable,int proto)2248 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2249 struct udphdr *uh,
2250 __be32 saddr, __be32 daddr,
2251 struct udp_table *udptable,
2252 int proto)
2253 {
2254 struct sock *sk, *first = NULL;
2255 unsigned short hnum = ntohs(uh->dest);
2256 struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2257 unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2258 unsigned int offset = offsetof(typeof(*sk), sk_node);
2259 int dif = skb->dev->ifindex;
2260 int sdif = inet_sdif(skb);
2261 struct hlist_node *node;
2262 struct sk_buff *nskb;
2263
2264 if (use_hash2) {
2265 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2266 udptable->mask;
2267 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2268 start_lookup:
2269 hslot = &udptable->hash2[hash2];
2270 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2271 }
2272
2273 sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2274 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2275 uh->source, saddr, dif, sdif, hnum))
2276 continue;
2277
2278 if (!first) {
2279 first = sk;
2280 continue;
2281 }
2282 nskb = skb_clone(skb, GFP_ATOMIC);
2283
2284 if (unlikely(!nskb)) {
2285 atomic_inc(&sk->sk_drops);
2286 __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2287 IS_UDPLITE(sk));
2288 __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2289 IS_UDPLITE(sk));
2290 continue;
2291 }
2292 if (udp_queue_rcv_skb(sk, nskb) > 0)
2293 consume_skb(nskb);
2294 }
2295
2296 /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2297 if (use_hash2 && hash2 != hash2_any) {
2298 hash2 = hash2_any;
2299 goto start_lookup;
2300 }
2301
2302 if (first) {
2303 if (udp_queue_rcv_skb(first, skb) > 0)
2304 consume_skb(skb);
2305 } else {
2306 kfree_skb(skb);
2307 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2308 proto == IPPROTO_UDPLITE);
2309 }
2310 return 0;
2311 }
2312
2313 /* Initialize UDP checksum. If exited with zero value (success),
2314 * CHECKSUM_UNNECESSARY means, that no more checks are required.
2315 * Otherwise, csum completion requires checksumming packet body,
2316 * including udp header and folding it to skb->csum.
2317 */
udp4_csum_init(struct sk_buff * skb,struct udphdr * uh,int proto)2318 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2319 int proto)
2320 {
2321 int err;
2322
2323 UDP_SKB_CB(skb)->partial_cov = 0;
2324 UDP_SKB_CB(skb)->cscov = skb->len;
2325
2326 if (proto == IPPROTO_UDPLITE) {
2327 err = udplite_checksum_init(skb, uh);
2328 if (err)
2329 return err;
2330
2331 if (UDP_SKB_CB(skb)->partial_cov) {
2332 skb->csum = inet_compute_pseudo(skb, proto);
2333 return 0;
2334 }
2335 }
2336
2337 /* Note, we are only interested in != 0 or == 0, thus the
2338 * force to int.
2339 */
2340 err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2341 inet_compute_pseudo);
2342 if (err)
2343 return err;
2344
2345 if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2346 /* If SW calculated the value, we know it's bad */
2347 if (skb->csum_complete_sw)
2348 return 1;
2349
2350 /* HW says the value is bad. Let's validate that.
2351 * skb->csum is no longer the full packet checksum,
2352 * so don't treat it as such.
2353 */
2354 skb_checksum_complete_unset(skb);
2355 }
2356
2357 return 0;
2358 }
2359
2360 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2361 * return code conversion for ip layer consumption
2362 */
udp_unicast_rcv_skb(struct sock * sk,struct sk_buff * skb,struct udphdr * uh)2363 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2364 struct udphdr *uh)
2365 {
2366 int ret;
2367
2368 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2369 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2370
2371 ret = udp_queue_rcv_skb(sk, skb);
2372
2373 /* a return value > 0 means to resubmit the input, but
2374 * it wants the return to be -protocol, or 0
2375 */
2376 if (ret > 0)
2377 return -ret;
2378 return 0;
2379 }
2380
2381 /*
2382 * All we need to do is get the socket, and then do a checksum.
2383 */
2384
__udp4_lib_rcv(struct sk_buff * skb,struct udp_table * udptable,int proto)2385 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2386 int proto)
2387 {
2388 struct sock *sk = NULL;
2389 struct udphdr *uh;
2390 unsigned short ulen;
2391 struct rtable *rt = skb_rtable(skb);
2392 __be32 saddr, daddr;
2393 struct net *net = dev_net(skb->dev);
2394 bool refcounted;
2395 int drop_reason;
2396
2397 drop_reason = SKB_DROP_REASON_NOT_SPECIFIED;
2398
2399 /*
2400 * Validate the packet.
2401 */
2402 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2403 goto drop; /* No space for header. */
2404
2405 uh = udp_hdr(skb);
2406 ulen = ntohs(uh->len);
2407 saddr = ip_hdr(skb)->saddr;
2408 daddr = ip_hdr(skb)->daddr;
2409
2410 if (ulen > skb->len)
2411 goto short_packet;
2412
2413 if (proto == IPPROTO_UDP) {
2414 /* UDP validates ulen. */
2415 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2416 goto short_packet;
2417 uh = udp_hdr(skb);
2418 }
2419
2420 if (udp4_csum_init(skb, uh, proto))
2421 goto csum_error;
2422
2423 sk = inet_steal_sock(net, skb, sizeof(struct udphdr), saddr, uh->source, daddr, uh->dest,
2424 &refcounted, udp_ehashfn);
2425 if (IS_ERR(sk))
2426 goto no_sk;
2427
2428 if (sk) {
2429 struct dst_entry *dst = skb_dst(skb);
2430 int ret;
2431
2432 if (unlikely(rcu_dereference(sk->sk_rx_dst) != dst))
2433 udp_sk_rx_dst_set(sk, dst);
2434
2435 ret = udp_unicast_rcv_skb(sk, skb, uh);
2436 if (refcounted)
2437 sock_put(sk);
2438 return ret;
2439 }
2440
2441 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2442 return __udp4_lib_mcast_deliver(net, skb, uh,
2443 saddr, daddr, udptable, proto);
2444
2445 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2446 if (sk)
2447 return udp_unicast_rcv_skb(sk, skb, uh);
2448 no_sk:
2449 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2450 goto drop;
2451 nf_reset_ct(skb);
2452
2453 /* No socket. Drop packet silently, if checksum is wrong */
2454 if (udp_lib_checksum_complete(skb))
2455 goto csum_error;
2456
2457 drop_reason = SKB_DROP_REASON_NO_SOCKET;
2458 __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2459 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2460
2461 /*
2462 * Hmm. We got an UDP packet to a port to which we
2463 * don't wanna listen. Ignore it.
2464 */
2465 sk_skb_reason_drop(sk, skb, drop_reason);
2466 return 0;
2467
2468 short_packet:
2469 drop_reason = SKB_DROP_REASON_PKT_TOO_SMALL;
2470 net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2471 proto == IPPROTO_UDPLITE ? "Lite" : "",
2472 &saddr, ntohs(uh->source),
2473 ulen, skb->len,
2474 &daddr, ntohs(uh->dest));
2475 goto drop;
2476
2477 csum_error:
2478 /*
2479 * RFC1122: OK. Discards the bad packet silently (as far as
2480 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2481 */
2482 drop_reason = SKB_DROP_REASON_UDP_CSUM;
2483 net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2484 proto == IPPROTO_UDPLITE ? "Lite" : "",
2485 &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2486 ulen);
2487 __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2488 drop:
2489 __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2490 sk_skb_reason_drop(sk, skb, drop_reason);
2491 return 0;
2492 }
2493
2494 /* We can only early demux multicast if there is a single matching socket.
2495 * If more than one socket found returns NULL
2496 */
__udp4_lib_mcast_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2497 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2498 __be16 loc_port, __be32 loc_addr,
2499 __be16 rmt_port, __be32 rmt_addr,
2500 int dif, int sdif)
2501 {
2502 struct udp_table *udptable = net->ipv4.udp_table;
2503 unsigned short hnum = ntohs(loc_port);
2504 struct sock *sk, *result;
2505 struct udp_hslot *hslot;
2506 unsigned int slot;
2507
2508 slot = udp_hashfn(net, hnum, udptable->mask);
2509 hslot = &udptable->hash[slot];
2510
2511 /* Do not bother scanning a too big list */
2512 if (hslot->count > 10)
2513 return NULL;
2514
2515 result = NULL;
2516 sk_for_each_rcu(sk, &hslot->head) {
2517 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2518 rmt_port, rmt_addr, dif, sdif, hnum)) {
2519 if (result)
2520 return NULL;
2521 result = sk;
2522 }
2523 }
2524
2525 return result;
2526 }
2527
2528 /* For unicast we should only early demux connected sockets or we can
2529 * break forwarding setups. The chains here can be long so only check
2530 * if the first socket is an exact match and if not move on.
2531 */
__udp4_lib_demux_lookup(struct net * net,__be16 loc_port,__be32 loc_addr,__be16 rmt_port,__be32 rmt_addr,int dif,int sdif)2532 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2533 __be16 loc_port, __be32 loc_addr,
2534 __be16 rmt_port, __be32 rmt_addr,
2535 int dif, int sdif)
2536 {
2537 struct udp_table *udptable = net->ipv4.udp_table;
2538 INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2539 unsigned short hnum = ntohs(loc_port);
2540 unsigned int hash2, slot2;
2541 struct udp_hslot *hslot2;
2542 __portpair ports;
2543 struct sock *sk;
2544
2545 hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2546 slot2 = hash2 & udptable->mask;
2547 hslot2 = &udptable->hash2[slot2];
2548 ports = INET_COMBINED_PORTS(rmt_port, hnum);
2549
2550 udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2551 if (inet_match(net, sk, acookie, ports, dif, sdif))
2552 return sk;
2553 /* Only check first socket in chain */
2554 break;
2555 }
2556 return NULL;
2557 }
2558
udp_v4_early_demux(struct sk_buff * skb)2559 int udp_v4_early_demux(struct sk_buff *skb)
2560 {
2561 struct net *net = dev_net(skb->dev);
2562 struct in_device *in_dev = NULL;
2563 const struct iphdr *iph;
2564 const struct udphdr *uh;
2565 struct sock *sk = NULL;
2566 struct dst_entry *dst;
2567 int dif = skb->dev->ifindex;
2568 int sdif = inet_sdif(skb);
2569 int ours;
2570
2571 /* validate the packet */
2572 if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2573 return 0;
2574
2575 iph = ip_hdr(skb);
2576 uh = udp_hdr(skb);
2577
2578 if (skb->pkt_type == PACKET_MULTICAST) {
2579 in_dev = __in_dev_get_rcu(skb->dev);
2580
2581 if (!in_dev)
2582 return 0;
2583
2584 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2585 iph->protocol);
2586 if (!ours)
2587 return 0;
2588
2589 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2590 uh->source, iph->saddr,
2591 dif, sdif);
2592 } else if (skb->pkt_type == PACKET_HOST) {
2593 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2594 uh->source, iph->saddr, dif, sdif);
2595 }
2596
2597 if (!sk)
2598 return 0;
2599
2600 skb->sk = sk;
2601 DEBUG_NET_WARN_ON_ONCE(sk_is_refcounted(sk));
2602 skb->destructor = sock_pfree;
2603 dst = rcu_dereference(sk->sk_rx_dst);
2604
2605 if (dst)
2606 dst = dst_check(dst, 0);
2607 if (dst) {
2608 u32 itag = 0;
2609
2610 /* set noref for now.
2611 * any place which wants to hold dst has to call
2612 * dst_hold_safe()
2613 */
2614 skb_dst_set_noref(skb, dst);
2615
2616 /* for unconnected multicast sockets we need to validate
2617 * the source on each packet
2618 */
2619 if (!inet_sk(sk)->inet_daddr && in_dev)
2620 return ip_mc_validate_source(skb, iph->daddr,
2621 iph->saddr,
2622 iph->tos & INET_DSCP_MASK,
2623 skb->dev, in_dev, &itag);
2624 }
2625 return 0;
2626 }
2627
udp_rcv(struct sk_buff * skb)2628 int udp_rcv(struct sk_buff *skb)
2629 {
2630 return __udp4_lib_rcv(skb, dev_net(skb->dev)->ipv4.udp_table, IPPROTO_UDP);
2631 }
2632
udp_destroy_sock(struct sock * sk)2633 void udp_destroy_sock(struct sock *sk)
2634 {
2635 struct udp_sock *up = udp_sk(sk);
2636 bool slow = lock_sock_fast(sk);
2637
2638 /* protects from races with udp_abort() */
2639 sock_set_flag(sk, SOCK_DEAD);
2640 udp_flush_pending_frames(sk);
2641 unlock_sock_fast(sk, slow);
2642 if (static_branch_unlikely(&udp_encap_needed_key)) {
2643 if (up->encap_type) {
2644 void (*encap_destroy)(struct sock *sk);
2645 encap_destroy = READ_ONCE(up->encap_destroy);
2646 if (encap_destroy)
2647 encap_destroy(sk);
2648 }
2649 if (udp_test_bit(ENCAP_ENABLED, sk))
2650 static_branch_dec(&udp_encap_needed_key);
2651 }
2652 }
2653
set_xfrm_gro_udp_encap_rcv(__u16 encap_type,unsigned short family,struct sock * sk)2654 static void set_xfrm_gro_udp_encap_rcv(__u16 encap_type, unsigned short family,
2655 struct sock *sk)
2656 {
2657 #ifdef CONFIG_XFRM
2658 if (udp_test_bit(GRO_ENABLED, sk) && encap_type == UDP_ENCAP_ESPINUDP) {
2659 if (family == AF_INET)
2660 WRITE_ONCE(udp_sk(sk)->gro_receive, xfrm4_gro_udp_encap_rcv);
2661 else if (IS_ENABLED(CONFIG_IPV6) && family == AF_INET6)
2662 WRITE_ONCE(udp_sk(sk)->gro_receive, ipv6_stub->xfrm6_gro_udp_encap_rcv);
2663 }
2664 #endif
2665 }
2666
2667 /*
2668 * Socket option code for UDP
2669 */
udp_lib_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen,int (* push_pending_frames)(struct sock *))2670 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2671 sockptr_t optval, unsigned int optlen,
2672 int (*push_pending_frames)(struct sock *))
2673 {
2674 struct udp_sock *up = udp_sk(sk);
2675 int val, valbool;
2676 int err = 0;
2677 int is_udplite = IS_UDPLITE(sk);
2678
2679 if (level == SOL_SOCKET) {
2680 err = sk_setsockopt(sk, level, optname, optval, optlen);
2681
2682 if (optname == SO_RCVBUF || optname == SO_RCVBUFFORCE) {
2683 sockopt_lock_sock(sk);
2684 /* paired with READ_ONCE in udp_rmem_release() */
2685 WRITE_ONCE(up->forward_threshold, sk->sk_rcvbuf >> 2);
2686 sockopt_release_sock(sk);
2687 }
2688 return err;
2689 }
2690
2691 if (optlen < sizeof(int))
2692 return -EINVAL;
2693
2694 if (copy_from_sockptr(&val, optval, sizeof(val)))
2695 return -EFAULT;
2696
2697 valbool = val ? 1 : 0;
2698
2699 switch (optname) {
2700 case UDP_CORK:
2701 if (val != 0) {
2702 udp_set_bit(CORK, sk);
2703 } else {
2704 udp_clear_bit(CORK, sk);
2705 lock_sock(sk);
2706 push_pending_frames(sk);
2707 release_sock(sk);
2708 }
2709 break;
2710
2711 case UDP_ENCAP:
2712 switch (val) {
2713 case 0:
2714 #ifdef CONFIG_XFRM
2715 case UDP_ENCAP_ESPINUDP:
2716 set_xfrm_gro_udp_encap_rcv(val, sk->sk_family, sk);
2717 #if IS_ENABLED(CONFIG_IPV6)
2718 if (sk->sk_family == AF_INET6)
2719 WRITE_ONCE(up->encap_rcv,
2720 ipv6_stub->xfrm6_udp_encap_rcv);
2721 else
2722 #endif
2723 WRITE_ONCE(up->encap_rcv,
2724 xfrm4_udp_encap_rcv);
2725 #endif
2726 fallthrough;
2727 case UDP_ENCAP_L2TPINUDP:
2728 WRITE_ONCE(up->encap_type, val);
2729 udp_tunnel_encap_enable(sk);
2730 break;
2731 default:
2732 err = -ENOPROTOOPT;
2733 break;
2734 }
2735 break;
2736
2737 case UDP_NO_CHECK6_TX:
2738 udp_set_no_check6_tx(sk, valbool);
2739 break;
2740
2741 case UDP_NO_CHECK6_RX:
2742 udp_set_no_check6_rx(sk, valbool);
2743 break;
2744
2745 case UDP_SEGMENT:
2746 if (val < 0 || val > USHRT_MAX)
2747 return -EINVAL;
2748 WRITE_ONCE(up->gso_size, val);
2749 break;
2750
2751 case UDP_GRO:
2752
2753 /* when enabling GRO, accept the related GSO packet type */
2754 if (valbool)
2755 udp_tunnel_encap_enable(sk);
2756 udp_assign_bit(GRO_ENABLED, sk, valbool);
2757 udp_assign_bit(ACCEPT_L4, sk, valbool);
2758 set_xfrm_gro_udp_encap_rcv(up->encap_type, sk->sk_family, sk);
2759 break;
2760
2761 /*
2762 * UDP-Lite's partial checksum coverage (RFC 3828).
2763 */
2764 /* The sender sets actual checksum coverage length via this option.
2765 * The case coverage > packet length is handled by send module. */
2766 case UDPLITE_SEND_CSCOV:
2767 if (!is_udplite) /* Disable the option on UDP sockets */
2768 return -ENOPROTOOPT;
2769 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2770 val = 8;
2771 else if (val > USHRT_MAX)
2772 val = USHRT_MAX;
2773 WRITE_ONCE(up->pcslen, val);
2774 udp_set_bit(UDPLITE_SEND_CC, sk);
2775 break;
2776
2777 /* The receiver specifies a minimum checksum coverage value. To make
2778 * sense, this should be set to at least 8 (as done below). If zero is
2779 * used, this again means full checksum coverage. */
2780 case UDPLITE_RECV_CSCOV:
2781 if (!is_udplite) /* Disable the option on UDP sockets */
2782 return -ENOPROTOOPT;
2783 if (val != 0 && val < 8) /* Avoid silly minimal values. */
2784 val = 8;
2785 else if (val > USHRT_MAX)
2786 val = USHRT_MAX;
2787 WRITE_ONCE(up->pcrlen, val);
2788 udp_set_bit(UDPLITE_RECV_CC, sk);
2789 break;
2790
2791 default:
2792 err = -ENOPROTOOPT;
2793 break;
2794 }
2795
2796 return err;
2797 }
2798 EXPORT_SYMBOL(udp_lib_setsockopt);
2799
udp_setsockopt(struct sock * sk,int level,int optname,sockptr_t optval,unsigned int optlen)2800 int udp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
2801 unsigned int optlen)
2802 {
2803 if (level == SOL_UDP || level == SOL_UDPLITE || level == SOL_SOCKET)
2804 return udp_lib_setsockopt(sk, level, optname,
2805 optval, optlen,
2806 udp_push_pending_frames);
2807 return ip_setsockopt(sk, level, optname, optval, optlen);
2808 }
2809
udp_lib_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)2810 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2811 char __user *optval, int __user *optlen)
2812 {
2813 struct udp_sock *up = udp_sk(sk);
2814 int val, len;
2815
2816 if (get_user(len, optlen))
2817 return -EFAULT;
2818
2819 if (len < 0)
2820 return -EINVAL;
2821
2822 len = min_t(unsigned int, len, sizeof(int));
2823
2824 switch (optname) {
2825 case UDP_CORK:
2826 val = udp_test_bit(CORK, sk);
2827 break;
2828
2829 case UDP_ENCAP:
2830 val = READ_ONCE(up->encap_type);
2831 break;
2832
2833 case UDP_NO_CHECK6_TX:
2834 val = udp_get_no_check6_tx(sk);
2835 break;
2836
2837 case UDP_NO_CHECK6_RX:
2838 val = udp_get_no_check6_rx(sk);
2839 break;
2840
2841 case UDP_SEGMENT:
2842 val = READ_ONCE(up->gso_size);
2843 break;
2844
2845 case UDP_GRO:
2846 val = udp_test_bit(GRO_ENABLED, sk);
2847 break;
2848
2849 /* The following two cannot be changed on UDP sockets, the return is
2850 * always 0 (which corresponds to the full checksum coverage of UDP). */
2851 case UDPLITE_SEND_CSCOV:
2852 val = READ_ONCE(up->pcslen);
2853 break;
2854
2855 case UDPLITE_RECV_CSCOV:
2856 val = READ_ONCE(up->pcrlen);
2857 break;
2858
2859 default:
2860 return -ENOPROTOOPT;
2861 }
2862
2863 if (put_user(len, optlen))
2864 return -EFAULT;
2865 if (copy_to_user(optval, &val, len))
2866 return -EFAULT;
2867 return 0;
2868 }
2869 EXPORT_SYMBOL(udp_lib_getsockopt);
2870
udp_getsockopt(struct sock * sk,int level,int optname,char __user * optval,int __user * optlen)2871 int udp_getsockopt(struct sock *sk, int level, int optname,
2872 char __user *optval, int __user *optlen)
2873 {
2874 if (level == SOL_UDP || level == SOL_UDPLITE)
2875 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2876 return ip_getsockopt(sk, level, optname, optval, optlen);
2877 }
2878
2879 /**
2880 * udp_poll - wait for a UDP event.
2881 * @file: - file struct
2882 * @sock: - socket
2883 * @wait: - poll table
2884 *
2885 * This is same as datagram poll, except for the special case of
2886 * blocking sockets. If application is using a blocking fd
2887 * and a packet with checksum error is in the queue;
2888 * then it could get return from select indicating data available
2889 * but then block when reading it. Add special case code
2890 * to work around these arguably broken applications.
2891 */
udp_poll(struct file * file,struct socket * sock,poll_table * wait)2892 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2893 {
2894 __poll_t mask = datagram_poll(file, sock, wait);
2895 struct sock *sk = sock->sk;
2896
2897 if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2898 mask |= EPOLLIN | EPOLLRDNORM;
2899
2900 /* Check for false positives due to checksum errors */
2901 if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2902 !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2903 mask &= ~(EPOLLIN | EPOLLRDNORM);
2904
2905 /* psock ingress_msg queue should not contain any bad checksum frames */
2906 if (sk_is_readable(sk))
2907 mask |= EPOLLIN | EPOLLRDNORM;
2908 return mask;
2909
2910 }
2911 EXPORT_SYMBOL(udp_poll);
2912
udp_abort(struct sock * sk,int err)2913 int udp_abort(struct sock *sk, int err)
2914 {
2915 if (!has_current_bpf_ctx())
2916 lock_sock(sk);
2917
2918 /* udp{v6}_destroy_sock() sets it under the sk lock, avoid racing
2919 * with close()
2920 */
2921 if (sock_flag(sk, SOCK_DEAD))
2922 goto out;
2923
2924 sk->sk_err = err;
2925 sk_error_report(sk);
2926 __udp_disconnect(sk, 0);
2927
2928 out:
2929 if (!has_current_bpf_ctx())
2930 release_sock(sk);
2931
2932 return 0;
2933 }
2934 EXPORT_SYMBOL_GPL(udp_abort);
2935
2936 struct proto udp_prot = {
2937 .name = "UDP",
2938 .owner = THIS_MODULE,
2939 .close = udp_lib_close,
2940 .pre_connect = udp_pre_connect,
2941 .connect = ip4_datagram_connect,
2942 .disconnect = udp_disconnect,
2943 .ioctl = udp_ioctl,
2944 .init = udp_init_sock,
2945 .destroy = udp_destroy_sock,
2946 .setsockopt = udp_setsockopt,
2947 .getsockopt = udp_getsockopt,
2948 .sendmsg = udp_sendmsg,
2949 .recvmsg = udp_recvmsg,
2950 .splice_eof = udp_splice_eof,
2951 .release_cb = ip4_datagram_release_cb,
2952 .hash = udp_lib_hash,
2953 .unhash = udp_lib_unhash,
2954 .rehash = udp_v4_rehash,
2955 .get_port = udp_v4_get_port,
2956 .put_port = udp_lib_unhash,
2957 #ifdef CONFIG_BPF_SYSCALL
2958 .psock_update_sk_prot = udp_bpf_update_proto,
2959 #endif
2960 .memory_allocated = &udp_memory_allocated,
2961 .per_cpu_fw_alloc = &udp_memory_per_cpu_fw_alloc,
2962
2963 .sysctl_mem = sysctl_udp_mem,
2964 .sysctl_wmem_offset = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2965 .sysctl_rmem_offset = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2966 .obj_size = sizeof(struct udp_sock),
2967 .h.udp_table = NULL,
2968 .diag_destroy = udp_abort,
2969 };
2970 EXPORT_SYMBOL(udp_prot);
2971
2972 /* ------------------------------------------------------------------------ */
2973 #ifdef CONFIG_PROC_FS
2974
2975 static unsigned short seq_file_family(const struct seq_file *seq);
seq_sk_match(struct seq_file * seq,const struct sock * sk)2976 static bool seq_sk_match(struct seq_file *seq, const struct sock *sk)
2977 {
2978 unsigned short family = seq_file_family(seq);
2979
2980 /* AF_UNSPEC is used as a match all */
2981 return ((family == AF_UNSPEC || family == sk->sk_family) &&
2982 net_eq(sock_net(sk), seq_file_net(seq)));
2983 }
2984
2985 #ifdef CONFIG_BPF_SYSCALL
2986 static const struct seq_operations bpf_iter_udp_seq_ops;
2987 #endif
udp_get_table_seq(struct seq_file * seq,struct net * net)2988 static struct udp_table *udp_get_table_seq(struct seq_file *seq,
2989 struct net *net)
2990 {
2991 const struct udp_seq_afinfo *afinfo;
2992
2993 #ifdef CONFIG_BPF_SYSCALL
2994 if (seq->op == &bpf_iter_udp_seq_ops)
2995 return net->ipv4.udp_table;
2996 #endif
2997
2998 afinfo = pde_data(file_inode(seq->file));
2999 return afinfo->udp_table ? : net->ipv4.udp_table;
3000 }
3001
udp_get_first(struct seq_file * seq,int start)3002 static struct sock *udp_get_first(struct seq_file *seq, int start)
3003 {
3004 struct udp_iter_state *state = seq->private;
3005 struct net *net = seq_file_net(seq);
3006 struct udp_table *udptable;
3007 struct sock *sk;
3008
3009 udptable = udp_get_table_seq(seq, net);
3010
3011 for (state->bucket = start; state->bucket <= udptable->mask;
3012 ++state->bucket) {
3013 struct udp_hslot *hslot = &udptable->hash[state->bucket];
3014
3015 if (hlist_empty(&hslot->head))
3016 continue;
3017
3018 spin_lock_bh(&hslot->lock);
3019 sk_for_each(sk, &hslot->head) {
3020 if (seq_sk_match(seq, sk))
3021 goto found;
3022 }
3023 spin_unlock_bh(&hslot->lock);
3024 }
3025 sk = NULL;
3026 found:
3027 return sk;
3028 }
3029
udp_get_next(struct seq_file * seq,struct sock * sk)3030 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
3031 {
3032 struct udp_iter_state *state = seq->private;
3033 struct net *net = seq_file_net(seq);
3034 struct udp_table *udptable;
3035
3036 do {
3037 sk = sk_next(sk);
3038 } while (sk && !seq_sk_match(seq, sk));
3039
3040 if (!sk) {
3041 udptable = udp_get_table_seq(seq, net);
3042
3043 if (state->bucket <= udptable->mask)
3044 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3045
3046 return udp_get_first(seq, state->bucket + 1);
3047 }
3048 return sk;
3049 }
3050
udp_get_idx(struct seq_file * seq,loff_t pos)3051 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
3052 {
3053 struct sock *sk = udp_get_first(seq, 0);
3054
3055 if (sk)
3056 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
3057 --pos;
3058 return pos ? NULL : sk;
3059 }
3060
udp_seq_start(struct seq_file * seq,loff_t * pos)3061 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
3062 {
3063 struct udp_iter_state *state = seq->private;
3064 state->bucket = MAX_UDP_PORTS;
3065
3066 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
3067 }
3068 EXPORT_SYMBOL(udp_seq_start);
3069
udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3070 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3071 {
3072 struct sock *sk;
3073
3074 if (v == SEQ_START_TOKEN)
3075 sk = udp_get_idx(seq, 0);
3076 else
3077 sk = udp_get_next(seq, v);
3078
3079 ++*pos;
3080 return sk;
3081 }
3082 EXPORT_SYMBOL(udp_seq_next);
3083
udp_seq_stop(struct seq_file * seq,void * v)3084 void udp_seq_stop(struct seq_file *seq, void *v)
3085 {
3086 struct udp_iter_state *state = seq->private;
3087 struct udp_table *udptable;
3088
3089 udptable = udp_get_table_seq(seq, seq_file_net(seq));
3090
3091 if (state->bucket <= udptable->mask)
3092 spin_unlock_bh(&udptable->hash[state->bucket].lock);
3093 }
3094 EXPORT_SYMBOL(udp_seq_stop);
3095
3096 /* ------------------------------------------------------------------------ */
udp4_format_sock(struct sock * sp,struct seq_file * f,int bucket)3097 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
3098 int bucket)
3099 {
3100 struct inet_sock *inet = inet_sk(sp);
3101 __be32 dest = inet->inet_daddr;
3102 __be32 src = inet->inet_rcv_saddr;
3103 __u16 destp = ntohs(inet->inet_dport);
3104 __u16 srcp = ntohs(inet->inet_sport);
3105
3106 seq_printf(f, "%5d: %08X:%04X %08X:%04X"
3107 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
3108 bucket, src, srcp, dest, destp, sp->sk_state,
3109 sk_wmem_alloc_get(sp),
3110 udp_rqueue_get(sp),
3111 0, 0L, 0,
3112 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
3113 0, sock_i_ino(sp),
3114 refcount_read(&sp->sk_refcnt), sp,
3115 atomic_read(&sp->sk_drops));
3116 }
3117
udp4_seq_show(struct seq_file * seq,void * v)3118 int udp4_seq_show(struct seq_file *seq, void *v)
3119 {
3120 seq_setwidth(seq, 127);
3121 if (v == SEQ_START_TOKEN)
3122 seq_puts(seq, " sl local_address rem_address st tx_queue "
3123 "rx_queue tr tm->when retrnsmt uid timeout "
3124 "inode ref pointer drops");
3125 else {
3126 struct udp_iter_state *state = seq->private;
3127
3128 udp4_format_sock(v, seq, state->bucket);
3129 }
3130 seq_pad(seq, '\n');
3131 return 0;
3132 }
3133
3134 #ifdef CONFIG_BPF_SYSCALL
3135 struct bpf_iter__udp {
3136 __bpf_md_ptr(struct bpf_iter_meta *, meta);
3137 __bpf_md_ptr(struct udp_sock *, udp_sk);
3138 uid_t uid __aligned(8);
3139 int bucket __aligned(8);
3140 };
3141
3142 struct bpf_udp_iter_state {
3143 struct udp_iter_state state;
3144 unsigned int cur_sk;
3145 unsigned int end_sk;
3146 unsigned int max_sk;
3147 int offset;
3148 struct sock **batch;
3149 bool st_bucket_done;
3150 };
3151
3152 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3153 unsigned int new_batch_sz);
bpf_iter_udp_batch(struct seq_file * seq)3154 static struct sock *bpf_iter_udp_batch(struct seq_file *seq)
3155 {
3156 struct bpf_udp_iter_state *iter = seq->private;
3157 struct udp_iter_state *state = &iter->state;
3158 struct net *net = seq_file_net(seq);
3159 int resume_bucket, resume_offset;
3160 struct udp_table *udptable;
3161 unsigned int batch_sks = 0;
3162 bool resized = false;
3163 struct sock *sk;
3164
3165 resume_bucket = state->bucket;
3166 resume_offset = iter->offset;
3167
3168 /* The current batch is done, so advance the bucket. */
3169 if (iter->st_bucket_done)
3170 state->bucket++;
3171
3172 udptable = udp_get_table_seq(seq, net);
3173
3174 again:
3175 /* New batch for the next bucket.
3176 * Iterate over the hash table to find a bucket with sockets matching
3177 * the iterator attributes, and return the first matching socket from
3178 * the bucket. The remaining matched sockets from the bucket are batched
3179 * before releasing the bucket lock. This allows BPF programs that are
3180 * called in seq_show to acquire the bucket lock if needed.
3181 */
3182 iter->cur_sk = 0;
3183 iter->end_sk = 0;
3184 iter->st_bucket_done = false;
3185 batch_sks = 0;
3186
3187 for (; state->bucket <= udptable->mask; state->bucket++) {
3188 struct udp_hslot *hslot2 = &udptable->hash2[state->bucket];
3189
3190 if (hlist_empty(&hslot2->head))
3191 continue;
3192
3193 iter->offset = 0;
3194 spin_lock_bh(&hslot2->lock);
3195 udp_portaddr_for_each_entry(sk, &hslot2->head) {
3196 if (seq_sk_match(seq, sk)) {
3197 /* Resume from the last iterated socket at the
3198 * offset in the bucket before iterator was stopped.
3199 */
3200 if (state->bucket == resume_bucket &&
3201 iter->offset < resume_offset) {
3202 ++iter->offset;
3203 continue;
3204 }
3205 if (iter->end_sk < iter->max_sk) {
3206 sock_hold(sk);
3207 iter->batch[iter->end_sk++] = sk;
3208 }
3209 batch_sks++;
3210 }
3211 }
3212 spin_unlock_bh(&hslot2->lock);
3213
3214 if (iter->end_sk)
3215 break;
3216 }
3217
3218 /* All done: no batch made. */
3219 if (!iter->end_sk)
3220 return NULL;
3221
3222 if (iter->end_sk == batch_sks) {
3223 /* Batching is done for the current bucket; return the first
3224 * socket to be iterated from the batch.
3225 */
3226 iter->st_bucket_done = true;
3227 goto done;
3228 }
3229 if (!resized && !bpf_iter_udp_realloc_batch(iter, batch_sks * 3 / 2)) {
3230 resized = true;
3231 /* After allocating a larger batch, retry one more time to grab
3232 * the whole bucket.
3233 */
3234 goto again;
3235 }
3236 done:
3237 return iter->batch[0];
3238 }
3239
bpf_iter_udp_seq_next(struct seq_file * seq,void * v,loff_t * pos)3240 static void *bpf_iter_udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3241 {
3242 struct bpf_udp_iter_state *iter = seq->private;
3243 struct sock *sk;
3244
3245 /* Whenever seq_next() is called, the iter->cur_sk is
3246 * done with seq_show(), so unref the iter->cur_sk.
3247 */
3248 if (iter->cur_sk < iter->end_sk) {
3249 sock_put(iter->batch[iter->cur_sk++]);
3250 ++iter->offset;
3251 }
3252
3253 /* After updating iter->cur_sk, check if there are more sockets
3254 * available in the current bucket batch.
3255 */
3256 if (iter->cur_sk < iter->end_sk)
3257 sk = iter->batch[iter->cur_sk];
3258 else
3259 /* Prepare a new batch. */
3260 sk = bpf_iter_udp_batch(seq);
3261
3262 ++*pos;
3263 return sk;
3264 }
3265
bpf_iter_udp_seq_start(struct seq_file * seq,loff_t * pos)3266 static void *bpf_iter_udp_seq_start(struct seq_file *seq, loff_t *pos)
3267 {
3268 /* bpf iter does not support lseek, so it always
3269 * continue from where it was stop()-ped.
3270 */
3271 if (*pos)
3272 return bpf_iter_udp_batch(seq);
3273
3274 return SEQ_START_TOKEN;
3275 }
3276
udp_prog_seq_show(struct bpf_prog * prog,struct bpf_iter_meta * meta,struct udp_sock * udp_sk,uid_t uid,int bucket)3277 static int udp_prog_seq_show(struct bpf_prog *prog, struct bpf_iter_meta *meta,
3278 struct udp_sock *udp_sk, uid_t uid, int bucket)
3279 {
3280 struct bpf_iter__udp ctx;
3281
3282 meta->seq_num--; /* skip SEQ_START_TOKEN */
3283 ctx.meta = meta;
3284 ctx.udp_sk = udp_sk;
3285 ctx.uid = uid;
3286 ctx.bucket = bucket;
3287 return bpf_iter_run_prog(prog, &ctx);
3288 }
3289
bpf_iter_udp_seq_show(struct seq_file * seq,void * v)3290 static int bpf_iter_udp_seq_show(struct seq_file *seq, void *v)
3291 {
3292 struct udp_iter_state *state = seq->private;
3293 struct bpf_iter_meta meta;
3294 struct bpf_prog *prog;
3295 struct sock *sk = v;
3296 uid_t uid;
3297 int ret;
3298
3299 if (v == SEQ_START_TOKEN)
3300 return 0;
3301
3302 lock_sock(sk);
3303
3304 if (unlikely(sk_unhashed(sk))) {
3305 ret = SEQ_SKIP;
3306 goto unlock;
3307 }
3308
3309 uid = from_kuid_munged(seq_user_ns(seq), sock_i_uid(sk));
3310 meta.seq = seq;
3311 prog = bpf_iter_get_info(&meta, false);
3312 ret = udp_prog_seq_show(prog, &meta, v, uid, state->bucket);
3313
3314 unlock:
3315 release_sock(sk);
3316 return ret;
3317 }
3318
bpf_iter_udp_put_batch(struct bpf_udp_iter_state * iter)3319 static void bpf_iter_udp_put_batch(struct bpf_udp_iter_state *iter)
3320 {
3321 while (iter->cur_sk < iter->end_sk)
3322 sock_put(iter->batch[iter->cur_sk++]);
3323 }
3324
bpf_iter_udp_seq_stop(struct seq_file * seq,void * v)3325 static void bpf_iter_udp_seq_stop(struct seq_file *seq, void *v)
3326 {
3327 struct bpf_udp_iter_state *iter = seq->private;
3328 struct bpf_iter_meta meta;
3329 struct bpf_prog *prog;
3330
3331 if (!v) {
3332 meta.seq = seq;
3333 prog = bpf_iter_get_info(&meta, true);
3334 if (prog)
3335 (void)udp_prog_seq_show(prog, &meta, v, 0, 0);
3336 }
3337
3338 if (iter->cur_sk < iter->end_sk) {
3339 bpf_iter_udp_put_batch(iter);
3340 iter->st_bucket_done = false;
3341 }
3342 }
3343
3344 static const struct seq_operations bpf_iter_udp_seq_ops = {
3345 .start = bpf_iter_udp_seq_start,
3346 .next = bpf_iter_udp_seq_next,
3347 .stop = bpf_iter_udp_seq_stop,
3348 .show = bpf_iter_udp_seq_show,
3349 };
3350 #endif
3351
seq_file_family(const struct seq_file * seq)3352 static unsigned short seq_file_family(const struct seq_file *seq)
3353 {
3354 const struct udp_seq_afinfo *afinfo;
3355
3356 #ifdef CONFIG_BPF_SYSCALL
3357 /* BPF iterator: bpf programs to filter sockets. */
3358 if (seq->op == &bpf_iter_udp_seq_ops)
3359 return AF_UNSPEC;
3360 #endif
3361
3362 /* Proc fs iterator */
3363 afinfo = pde_data(file_inode(seq->file));
3364 return afinfo->family;
3365 }
3366
3367 const struct seq_operations udp_seq_ops = {
3368 .start = udp_seq_start,
3369 .next = udp_seq_next,
3370 .stop = udp_seq_stop,
3371 .show = udp4_seq_show,
3372 };
3373 EXPORT_SYMBOL(udp_seq_ops);
3374
3375 static struct udp_seq_afinfo udp4_seq_afinfo = {
3376 .family = AF_INET,
3377 .udp_table = NULL,
3378 };
3379
udp4_proc_init_net(struct net * net)3380 static int __net_init udp4_proc_init_net(struct net *net)
3381 {
3382 if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
3383 sizeof(struct udp_iter_state), &udp4_seq_afinfo))
3384 return -ENOMEM;
3385 return 0;
3386 }
3387
udp4_proc_exit_net(struct net * net)3388 static void __net_exit udp4_proc_exit_net(struct net *net)
3389 {
3390 remove_proc_entry("udp", net->proc_net);
3391 }
3392
3393 static struct pernet_operations udp4_net_ops = {
3394 .init = udp4_proc_init_net,
3395 .exit = udp4_proc_exit_net,
3396 };
3397
udp4_proc_init(void)3398 int __init udp4_proc_init(void)
3399 {
3400 return register_pernet_subsys(&udp4_net_ops);
3401 }
3402
udp4_proc_exit(void)3403 void udp4_proc_exit(void)
3404 {
3405 unregister_pernet_subsys(&udp4_net_ops);
3406 }
3407 #endif /* CONFIG_PROC_FS */
3408
3409 static __initdata unsigned long uhash_entries;
set_uhash_entries(char * str)3410 static int __init set_uhash_entries(char *str)
3411 {
3412 ssize_t ret;
3413
3414 if (!str)
3415 return 0;
3416
3417 ret = kstrtoul(str, 0, &uhash_entries);
3418 if (ret)
3419 return 0;
3420
3421 if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
3422 uhash_entries = UDP_HTABLE_SIZE_MIN;
3423 return 1;
3424 }
3425 __setup("uhash_entries=", set_uhash_entries);
3426
udp_table_init(struct udp_table * table,const char * name)3427 void __init udp_table_init(struct udp_table *table, const char *name)
3428 {
3429 unsigned int i;
3430
3431 table->hash = alloc_large_system_hash(name,
3432 2 * sizeof(struct udp_hslot),
3433 uhash_entries,
3434 21, /* one slot per 2 MB */
3435 0,
3436 &table->log,
3437 &table->mask,
3438 UDP_HTABLE_SIZE_MIN,
3439 UDP_HTABLE_SIZE_MAX);
3440
3441 table->hash2 = table->hash + (table->mask + 1);
3442 for (i = 0; i <= table->mask; i++) {
3443 INIT_HLIST_HEAD(&table->hash[i].head);
3444 table->hash[i].count = 0;
3445 spin_lock_init(&table->hash[i].lock);
3446 }
3447 for (i = 0; i <= table->mask; i++) {
3448 INIT_HLIST_HEAD(&table->hash2[i].head);
3449 table->hash2[i].count = 0;
3450 spin_lock_init(&table->hash2[i].lock);
3451 }
3452 }
3453
udp_flow_hashrnd(void)3454 u32 udp_flow_hashrnd(void)
3455 {
3456 static u32 hashrnd __read_mostly;
3457
3458 net_get_random_once(&hashrnd, sizeof(hashrnd));
3459
3460 return hashrnd;
3461 }
3462 EXPORT_SYMBOL(udp_flow_hashrnd);
3463
udp_sysctl_init(struct net * net)3464 static void __net_init udp_sysctl_init(struct net *net)
3465 {
3466 net->ipv4.sysctl_udp_rmem_min = PAGE_SIZE;
3467 net->ipv4.sysctl_udp_wmem_min = PAGE_SIZE;
3468
3469 #ifdef CONFIG_NET_L3_MASTER_DEV
3470 net->ipv4.sysctl_udp_l3mdev_accept = 0;
3471 #endif
3472 }
3473
udp_pernet_table_alloc(unsigned int hash_entries)3474 static struct udp_table __net_init *udp_pernet_table_alloc(unsigned int hash_entries)
3475 {
3476 struct udp_table *udptable;
3477 int i;
3478
3479 udptable = kmalloc(sizeof(*udptable), GFP_KERNEL);
3480 if (!udptable)
3481 goto out;
3482
3483 udptable->hash = vmalloc_huge(hash_entries * 2 * sizeof(struct udp_hslot),
3484 GFP_KERNEL_ACCOUNT);
3485 if (!udptable->hash)
3486 goto free_table;
3487
3488 udptable->hash2 = udptable->hash + hash_entries;
3489 udptable->mask = hash_entries - 1;
3490 udptable->log = ilog2(hash_entries);
3491
3492 for (i = 0; i < hash_entries; i++) {
3493 INIT_HLIST_HEAD(&udptable->hash[i].head);
3494 udptable->hash[i].count = 0;
3495 spin_lock_init(&udptable->hash[i].lock);
3496
3497 INIT_HLIST_HEAD(&udptable->hash2[i].head);
3498 udptable->hash2[i].count = 0;
3499 spin_lock_init(&udptable->hash2[i].lock);
3500 }
3501
3502 return udptable;
3503
3504 free_table:
3505 kfree(udptable);
3506 out:
3507 return NULL;
3508 }
3509
udp_pernet_table_free(struct net * net)3510 static void __net_exit udp_pernet_table_free(struct net *net)
3511 {
3512 struct udp_table *udptable = net->ipv4.udp_table;
3513
3514 if (udptable == &udp_table)
3515 return;
3516
3517 kvfree(udptable->hash);
3518 kfree(udptable);
3519 }
3520
udp_set_table(struct net * net)3521 static void __net_init udp_set_table(struct net *net)
3522 {
3523 struct udp_table *udptable;
3524 unsigned int hash_entries;
3525 struct net *old_net;
3526
3527 if (net_eq(net, &init_net))
3528 goto fallback;
3529
3530 old_net = current->nsproxy->net_ns;
3531 hash_entries = READ_ONCE(old_net->ipv4.sysctl_udp_child_hash_entries);
3532 if (!hash_entries)
3533 goto fallback;
3534
3535 /* Set min to keep the bitmap on stack in udp_lib_get_port() */
3536 if (hash_entries < UDP_HTABLE_SIZE_MIN_PERNET)
3537 hash_entries = UDP_HTABLE_SIZE_MIN_PERNET;
3538 else
3539 hash_entries = roundup_pow_of_two(hash_entries);
3540
3541 udptable = udp_pernet_table_alloc(hash_entries);
3542 if (udptable) {
3543 net->ipv4.udp_table = udptable;
3544 } else {
3545 pr_warn("Failed to allocate UDP hash table (entries: %u) "
3546 "for a netns, fallback to the global one\n",
3547 hash_entries);
3548 fallback:
3549 net->ipv4.udp_table = &udp_table;
3550 }
3551 }
3552
udp_pernet_init(struct net * net)3553 static int __net_init udp_pernet_init(struct net *net)
3554 {
3555 udp_sysctl_init(net);
3556 udp_set_table(net);
3557
3558 return 0;
3559 }
3560
udp_pernet_exit(struct net * net)3561 static void __net_exit udp_pernet_exit(struct net *net)
3562 {
3563 udp_pernet_table_free(net);
3564 }
3565
3566 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3567 .init = udp_pernet_init,
3568 .exit = udp_pernet_exit,
3569 };
3570
3571 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
DEFINE_BPF_ITER_FUNC(udp,struct bpf_iter_meta * meta,struct udp_sock * udp_sk,uid_t uid,int bucket)3572 DEFINE_BPF_ITER_FUNC(udp, struct bpf_iter_meta *meta,
3573 struct udp_sock *udp_sk, uid_t uid, int bucket)
3574
3575 static int bpf_iter_udp_realloc_batch(struct bpf_udp_iter_state *iter,
3576 unsigned int new_batch_sz)
3577 {
3578 struct sock **new_batch;
3579
3580 new_batch = kvmalloc_array(new_batch_sz, sizeof(*new_batch),
3581 GFP_USER | __GFP_NOWARN);
3582 if (!new_batch)
3583 return -ENOMEM;
3584
3585 bpf_iter_udp_put_batch(iter);
3586 kvfree(iter->batch);
3587 iter->batch = new_batch;
3588 iter->max_sk = new_batch_sz;
3589
3590 return 0;
3591 }
3592
3593 #define INIT_BATCH_SZ 16
3594
bpf_iter_init_udp(void * priv_data,struct bpf_iter_aux_info * aux)3595 static int bpf_iter_init_udp(void *priv_data, struct bpf_iter_aux_info *aux)
3596 {
3597 struct bpf_udp_iter_state *iter = priv_data;
3598 int ret;
3599
3600 ret = bpf_iter_init_seq_net(priv_data, aux);
3601 if (ret)
3602 return ret;
3603
3604 ret = bpf_iter_udp_realloc_batch(iter, INIT_BATCH_SZ);
3605 if (ret)
3606 bpf_iter_fini_seq_net(priv_data);
3607
3608 return ret;
3609 }
3610
bpf_iter_fini_udp(void * priv_data)3611 static void bpf_iter_fini_udp(void *priv_data)
3612 {
3613 struct bpf_udp_iter_state *iter = priv_data;
3614
3615 bpf_iter_fini_seq_net(priv_data);
3616 kvfree(iter->batch);
3617 }
3618
3619 static const struct bpf_iter_seq_info udp_seq_info = {
3620 .seq_ops = &bpf_iter_udp_seq_ops,
3621 .init_seq_private = bpf_iter_init_udp,
3622 .fini_seq_private = bpf_iter_fini_udp,
3623 .seq_priv_size = sizeof(struct bpf_udp_iter_state),
3624 };
3625
3626 static struct bpf_iter_reg udp_reg_info = {
3627 .target = "udp",
3628 .ctx_arg_info_size = 1,
3629 .ctx_arg_info = {
3630 { offsetof(struct bpf_iter__udp, udp_sk),
3631 PTR_TO_BTF_ID_OR_NULL | PTR_TRUSTED },
3632 },
3633 .seq_info = &udp_seq_info,
3634 };
3635
bpf_iter_register(void)3636 static void __init bpf_iter_register(void)
3637 {
3638 udp_reg_info.ctx_arg_info[0].btf_id = btf_sock_ids[BTF_SOCK_TYPE_UDP];
3639 if (bpf_iter_reg_target(&udp_reg_info))
3640 pr_warn("Warning: could not register bpf iterator udp\n");
3641 }
3642 #endif
3643
udp_init(void)3644 void __init udp_init(void)
3645 {
3646 unsigned long limit;
3647 unsigned int i;
3648
3649 udp_table_init(&udp_table, "UDP");
3650 limit = nr_free_buffer_pages() / 8;
3651 limit = max(limit, 128UL);
3652 sysctl_udp_mem[0] = limit / 4 * 3;
3653 sysctl_udp_mem[1] = limit;
3654 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3655
3656 /* 16 spinlocks per cpu */
3657 udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3658 udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3659 GFP_KERNEL);
3660 if (!udp_busylocks)
3661 panic("UDP: failed to alloc udp_busylocks\n");
3662 for (i = 0; i < (1U << udp_busylocks_log); i++)
3663 spin_lock_init(udp_busylocks + i);
3664
3665 if (register_pernet_subsys(&udp_sysctl_ops))
3666 panic("UDP: failed to init sysctl parameters.\n");
3667
3668 #if defined(CONFIG_BPF_SYSCALL) && defined(CONFIG_PROC_FS)
3669 bpf_iter_register();
3670 #endif
3671 }
3672